OpenAI Launches GPT-5.6 With Three Models: Sol, Terra and Luna

At three overlapping UN summits on artificial intelligence in Geneva, representatives from more than 190 countries watched China's industry minister appear virtually everywhere, while the United States limited its participation to a junior official.

Cao Liangliang, a Google DeepMind director who worked on Gemini, is ending his Silicon Valley career to return to Hong Kong as a professor. It's the latest high-profile departure from Google's AI lab amid an intensifying war for AI research talent.

Sam Altman has admitted he was wrong about the scale of AI-driven job losses, while Dario Amodei is backing away from earlier forecasts that AI would wipe out half of all office jobs. Meanwhile, the tech industry has shed 120,000 to 165,000 jobs since the start of 2026.

UN Secretary-General António Guterres demanded a binding treaty banning fully autonomous weapons by 2026 at the first Global Dialogue on AI Governance in Geneva. The United States opposes a categorical ban, despite support from more than 120 countries.

Microsoft laid off 136 people at id Software, cutting the team responsible for the id Tech engine down to a single position. Gamers are joking that AI will now take over the engine's development, while Xbox insists it still plans to keep using it despite the cuts.

The former U.S. Federal Reserve chairman and Nobel laureate has joined Anthropic's Long-Term Benefit Trust, the independent body overseeing the company's AI development. He is the first economist of this stature to sit on the oversight body of any major AI lab.

Elon Musk's company has completed its transition from the xAI brand to SpaceXAI and unveiled Grok 4.5, a model priced far below Anthropic's competing Opus and available immediately in the recently acquired Cursor.

Google DeepMind and Durham University have launched a new feature in Antigravity that lets historians analyze Roman and Greek inscriptions in plain English, without writing code.

OpenAI has introduced ChatGPT Work, an agent built on the GPT-5.6 model that independently plans and executes multi-hour office tasks for businesses, drawing on data from Slack, Gmail, Google Drive and CRM systems.

The city of Cheyenne, Wyoming halted acceptance of industrial wastewater from all local data centers after a contractor building Meta's campus contaminated the reclaimed water system with a rare, metal-resistant bacterium.

The combined IPOs of Anthropic, OpenAI, and SpaceX could top $4 trillion in value, more than all US venture-backed exit deals since 2000 combined.

London startup Marker, co-founded by Google DeepMind's former head of creative, has emerged from stealth with a $13 million seed round. Its product is an AI-powered text editor meant to support writers rather than write for them.

AlphaEvolve, an AI agent for algorithm optimization built on Gemini, has reached general availability on the Gemini Enterprise platform. BASF improved its planning models by more than 80 percent, while Klarna doubled the throughput of its ML pipeline.

Scientists at the Alan Turing Institute broke malicious requests into seemingly innocent steps of a normal coding workflow, achieving a 100 percent attack success rate against four models powering GitHub Copilot, even though the same requests asked directly in chat were refused almost every time.

Poland's Standing Committee of the Council of Ministers has adopted a resolution guaranteeing the country's participation in the EU's AI gigafactory initiative, pledging 100 million euros for the first phase of computing capacity procurement. It is the first concrete step after months of negotiations with Lithuania and other Baltic states.

Meta will begin production in September 2026 of the next generation of its in-house MTIA chip, developed with Broadcom and manufactured by TSMC, as part of a plan to double the company's computing power by 2027.

Wiz Research detailed a technique called GhostApproval that uses symbolic links to trick Amazon Q, Claude Code, Cursor, Augment, Google Antigravity and Windsurf into writing files outside the project sandbox without the developer's knowledge.

Researchers from Tel Aviv University, Technion, and Intuit have described a new attack technique that exploits AI model hallucinations to smuggle malicious code through popular coding tools, including GitHub Copilot and Cursor.

Allianz Partners, the travel insurance and assistance arm of insurance giant Allianz, will cut 1,500 to 1,800 jobs over 12 to 18 months. The reduction is driven by AI-powered automation of customer service and claims handling.

California startup PrismML, a Caltech spin-off, says it has compressed the 27-billion-parameter Qwen 3.6 model to run entirely locally on an iPhone 17 Pro without any loss in quality.

Poland's Ministry of Digital Affairs held a meeting on July 9 for companies interested in building an EU AI gigafactory in Poland, as negotiations over the InvestAI funding model near completion.

A KPMG survey of over two thousand executives finds companies struggling to explain rising AI costs after vendors shifted to usage-based billing. Uber, Amazon, Walmart, and Meta are already cutting internal AI tool budgets.

The Council and European Parliament approved the Omnibus VII package, giving companies more time to implement rules for high-risk AI systems while accelerating the requirement to label AI-generated content.

Google and Sony Pictures Networks India are partnering to bring the Gemini assistant to the new season of Kaun Banega Crorepati, India's version of Who Wants to Be a Millionaire, hosted for decades by Amitabh Bachchan.

A new World Bank report finds that AI-driven job automation could cut Poland's PIT and social security (ZUS) tax revenue by up to 1 percent of GDP at its peak. The institution suggests raising corporate tax (CIT), but Polish economists are pushing back.

Perplexity, valued at $20 billion, has been internally testing a coding agent called Teammate since May 2026, built to run entire software projects rather than just suggest code snippets. The company's CTO wants engineers to stop reading code themselves by the end of the year.

Meta's new Muse Image feature lets anyone tag a public Instagram profile to generate an AI image using that person's photos, without notifying the account owner. Turning it off requires manually digging into settings.

French AI lab Mistral AI has released Robostral Navigate, a model that lets robots navigate complex indoor and outdoor spaces using a single standard camera and natural-language commands, with no LiDAR or depth sensors required.

In just two weeks, Anthropic recruited Nobel laureate John Jumper, Berkeley computer science chair Jelani Nelson, and two leading Google Gemini researchers. Universities and Google DeepMind warn the AI talent race is entering a new phase.

Google DeepMind added background task execution, remote MCP server support, custom functions, and credential refresh to Managed Agents in the Gemini API, moving AI agents closer to production use.

Nvidia shares rose after reports that Beijing is preparing to conditionally allow Chinese tech firms to buy H200 chips. The under-200,000-unit cap is less than half of what Chinese companies had requested.

Nvidia shares have fallen 16 percent from their May peak, wiping out roughly $1 trillion in valuation and pulling the company back to pre-AI-boom levels. Capital is flowing instead to Micron, AMD and Intel, whose combined market value rose by about $2 trillion in the second quarter.

The New York Times, Daily News and other outlets have filed a motion seeking sanctions against OpenAI over alleged concealment and destruction of evidence in the lawsuit over the use of news articles to train ChatGPT. The litigation alone has already cost the Times more than $28 million.

OpenAI is rolling out GPT-Live-1 and GPT-Live-1 mini, new full-duplex voice models that listen and speak simultaneously and offer real-time simultaneous translation in ChatGPT.

China's industry ministry warned of a security vulnerability in Anthropic's Claude Code, and Anthropic confirmed that hidden code tracked users' locations, explaining it as an experiment against illegal model distillation.

Chinese startup DeepSeek is working on its own AI inference chip to cut its reliance on Nvidia and Huawei processors, Reuters reported, citing three people familiar with the matter. It's the latest move in the race for chip independence, following similar steps by OpenAI and Anthropic.

The Maven Smart System, built on Palantir's algorithms, is designed to track Russian unit movements in real time along the border from Finland to Romania, including near Poland. NATO has confirmed the system is fully operational, though Germany and France have voiced reservations about its use.

Illinois Governor JB Pritzker has signed the AI Safety Measures Act, requiring the largest AI model developers to undergo annual independent audits and report incidents within 72 hours. It is the first law of its kind in the US, and it has the backing of both OpenAI and Anthropic.

The US Commerce Department has lifted sudden export restrictions on Anthropic's Claude Fable 5 and Mythos 5 models, ending a two-week standoff triggered by reports of a jailbreak method. The company agreed to closer cooperation with the government on safety testing.

Warsaw startup Nomagic has deployed a vision-language-action model with its first customers that cuts human interventions on warehouse robots by roughly half. The new AI lab is led by Markus Wulfmeier, formerly a researcher on Google DeepMind's Gemini Robotics team.

Norm AI, maker of an agentic AI-driven law firm, raised $120 million in a Series C round led by Khosla Ventures, reaching a $1.2 billion valuation just three years after its founding.

Starting July 9, OpenAI is gradually opening access to three new models, Sol, Terra and Luna, after Sam Altman told employees the phased rollout stemmed from talks with the US administration rather than a purely business decision.

Prime Video removed the thriller Deadly Patient from its German catalog after viewers mocked the flat, AI-generated dubbing. It's the latest episode in a dispute over dubbing automation, following the earlier withdrawal of English-language AI dubs for anime.

Startup Prime Intellect closed a $130 million Series A round at a $1 billion valuation, offering companies a full stack of tools to build their own AI agents independent of major labs. Customers including Ramp and Zapier say the resulting agents are cheaper and more accurate than models from leading providers.

Anthropic researchers have described J-space, a small set of neural patterns in Claude that governs what the model can report and consciously control. The company stresses this isn't proof of consciousness, but a tool for detecting hidden goals and deception in the model.

A new Deloitte study finds that 50 percent of American parents worry their children rely too much on artificial intelligence, while schools struggle to set clear rules. In Poland, 43 percent of children aged 7 to 14 have already used ChatGPT.

As of July 15, China's first law regulating chatbots that mimic human personality has taken effect. ByteDance is shutting down personalized agents in Doubao, Alibaba is doing the same with Qwen, and users are losing access to months of relationships built with their AI companions.

An analysis of 61,200 decisions by six language models shows AI chatbots end stories negatively 13 to 17 times more often when the protagonist has schizophrenia, HIV, or AIDS than when they are healthy. Researchers warn that a third of American adults ask these same models about their health.

Nvidia and LangChain announced a joint blueprint called NemoClaw, pairing the open Nemotron 3 Ultra model with the Deep Agents framework. In benchmark tests it matched the quality of closed competitors at roughly one-tenth the inference cost.

Anthropic planned to end free access to its priciest model, Claude Fable 5, on July 8, but extended the deadline to July 12 after user backlash. From that date, usage beyond half the weekly limit will require paying for separate credits.

Cao Liangliang, a Google DeepMind director who worked on Gemini, is taking up an artificial intelligence chair at Hong Kong Polytechnic University after twenty years in the US, joining a growing wave of AI researchers leaving Silicon Valley for Asia.

SpaceXAI, formerly xAI, has launched Grok 4.5, the first model built in collaboration with Cursor following SpaceX's $60 billion acquisition of the company. The model is rolling out in Cursor and via API, but will initially skip users in the European Union.

Check Point Research reports that government institutions in Poland face over 3,100 cyberattacks per week, crossing that threshold for the first time. The firm points to generative AI as a growing channel for corporate data leaks.

All 193 UN member states met for the first time solely to discuss artificial intelligence. Secretary-General Guterres outlined four priorities and announced a follow-up round of talks in New York in 2027.

ESET's report for the first half of 2026 found that among nearly 900,000 scanned AI agent plugins, more than 3,000 turned out to be malicious, with the number of suspicious extensions multiplying within just a few months.

A new analysis of Poland's labor market shows that nearly 3.7 million people work in occupations most exposed to artificial intelligence, while International Monetary Fund forecasts point to as many as 5.5 million jobs at risk within a decade.

Microsoft has started replacing OpenAI and Anthropic models with its own MAI models in Excel, Outlook and Copilot to curb rising licensing costs, though the new models currently handle only a small share of queries.

Researchers at Wiz described a systemic vulnerability in six popular AI coding agents, including Cursor, Claude Code, and Amazon Q, that uses symlinks to overwrite sensitive files despite an apparent user approval.

Meta Superintelligence Labs released an early preview of Muse Video, a text-to-video model with native audio that ranked third in the Arena user preference leaderboard.

Researchers from Tel Aviv University, Technion, and Intuit have shown how registering AI-hallucinated repository names can be used to hijack nine popular coding tools, including Cursor and GitHub Copilot.

Demand for SK Hynix shares in its record Nasdaq offering exceeded the available pool more than sevenfold before pricing was even finalized. All proceeds will fund expansion of the HBM memory factories powering AI accelerators.

Allianz Partners will cut 1,500 to 1,800 jobs across Europe within 12 to 18 months as artificial intelligence takes over customer phone support and claims handling. The company's CEO openly admitted that automation is behind the cuts.

General Intuition, a startup led by Pim de Witte, has raised $320 million at a $2.3 billion valuation for an AI model trained on millions of hours of video game data that can control a real robot after just minutes of fine-tuning.

The European Parliament has approved the Omnibus VII package, pushing back some obligations for high-risk AI systems by up to a year and a half while accelerating rules on labeling AI-generated content and banning nudify-style apps.

Verity Harding, former head of public policy at Google DeepMind, tells WIRED that the United States' nationalist approach to artificial intelligence is proof the worst-case scenario for the technology is coming true.

A year ago, the heads of OpenAI, Anthropic, and Amazon were warning of mass layoffs driven by AI. Now they're talking about labor shortages and admitting they got it wrong, even as real layoffs keep hitting the tech industry.

Anthropic has launched a public beta of Claude Code and Claude Cowork for the US federal government as part of Claude for Government Desktop, running in a FedRAMP High-authorized environment. It marks the company's first major step toward bringing its coding agents and office tools to US public agencies.

Perplexity has been internally testing a tool codenamed Teammate since May, designed to run software projects from start to finish - a potential rival to Cursor and Claude Code, though the company won't confirm plans for a public release.

A new World Bank report estimates that the rise of artificial intelligence will shrink Poland's personal income tax and ZUS social security revenue by up to 0.8 percentage points of GDP by 2035, and Polish economists are already split over whether the fix should be a higher corporate tax rate.

Anthropic is expanding its office agent Claude Cowork with mobile and web versions, plus new tools for working with email, calendar, and files in Microsoft 365. The move comes as Microsoft makes a similar push, folding Cowork's technology into its own Copilot.

Researchers at the Alan Turing Institute found that GitHub Copilot refuses harmful requests in chat but carries them out without exception when the same requests are broken into ordinary coding steps. In a test of 816 attempts, the bypass succeeded 100 percent of the time.

Gartner forecasts that the share of organizations relying on small, few-person engineering teams will rise from 15 to 60 percent within three years. Analysts stress this is a restructuring of roles around working with AI agents, not a cost-cutting move, and warn against slashing junior hiring.

Google DeepMind scrapped the Gemini 2.5 Pro architecture and started a new training cycle for its successor, pushing the launch from June to July 17, 2026. The decision coincided with the departure of two top scientists to OpenAI and Anthropic and a drop in Alphabet's stock.

Backed by Intel and Microsoft, Syntiant, a maker of energy-efficient AI processors for wearables and automotive, has filed for a Nasdaq IPO under the ticker SYTN, aiming to raise up to $300 million.

JetBrains has launched AI for Teams and Organizations, a management layer that ties Claude Code, Codex and Gemini CLI together into a single cost- and permissions-control system for businesses.

China's Ministry of Industry and Information Technology says Anthropic's Claude Code sent user location and identity data without consent; Alibaba ordered employees to switch to its own Qoder tool starting July 10.

At the opening of the UN's first Global Dialogue on AI Governance in Geneva, António Guterres urged states to conclude a binding treaty banning lethal autonomous weapons by the end of 2026. 127 countries back a legal ban, while the US, UK, Russia and Israel prefer non-binding ethical frameworks.

The controversial digital character Tilly Norwood will play the lead role in the comedy-drama Misaligned. The SAG-AFTRA union calls her a program trained on stolen performances by real actors.

Meta's new Muse Image model lets anyone generate a photo of any public Instagram user just by tagging their username in a prompt. Accounts are opted in by default, and opting out has to be found manually in settings.

Researchers at Noma Security disclosed GitLost, a vulnerability that let anyone with no credentials or access extract private organization repository data through a simple GitHub issue processed by GitHub Agentic Workflows. GitHub has not yet shipped a fix.

Warsaw-based Nomagic says it's among the first in the world to run a vision-language-action model in live, paying-customer warehouses, cutting the need for human intervention in half. The company's AI lab is led by Markus Wulfmeier, a former Google DeepMind Gemini Robotics researcher.

OpenAI is rolling out GPT-Live, a voice technology that lets ChatGPT listen and speak at the same time, available from July 8, 2026 on iOS, Android, in the browser, and in Apple CarPlay.

The share of US companies using Chinese AI models on the OpenRouter platform has topped 30 percent weekly, with Z.ai's GLM-5.2 model costing up to nine times less than Claude Opus at comparable quality.

Illinois Governor JB Pritzker has signed the Artificial Intelligence Safety Measures Act, requiring large AI model developers to publish safety frameworks, report incidents within 72 hours, and undergo annual independent audits. Violations carry penalties of up to $3 million.

Starting Thursday, OpenAI is making its GPT-5.6 model family (Sol, Terra, Luna) available to the general public after the U.S. Department of Commerce completed a security review of the most powerful version. Previously, access to GPT-5.6 Sol was limited to around 20 trusted partners individually approved by the government.

A Surfshark analysis shows regulators and courts across Europe have fined seven major tech companies more than $3.5 billion since 2022 for illegally training AI on user data without consent, just as the EU AI Act's full penalty regime takes effect on August 2.

Sysdig researchers documented the first fully autonomous ransomware attack, in which an AI agent independently breached a network, exfiltrated data, and encrypted databases without a human at the keyboard.

The US AI data center construction boom has attracted organized crime rings specializing in cargo theft. Two intercepted trailers near Chicago alone yielded equipment and copper worth $1.3 million.

Meta has announced construction of its first data center in Canada, a gigawatt-scale complex in Sturgeon County, Alberta worth over $13 billion. It's part of the company's global AI infrastructure buildout, backed by planned 2026 capital spending of up to $145 billion.

Ukraine's Ministry of Digital Transformation says future AI procurement for government, business and the military will prioritize models that can run on domestic servers beyond any foreign provider's remote control. The policy follows a US government directive restricting access to some Anthropic models.

NATO is developing a digital surveillance system that links satellites, drones and ground sensors with artificial intelligence to track troop movements from Romania to Finland and speed up decisions on responding to threats.

Researchers at Tenet Security showed how a single crafted bug report in Sentry can force Claude Code, Cursor, and OpenAI Codex to execute malicious code on a developer's machine. More than 2,300 organizations worldwide were found vulnerable.

Venice, a privacy-focused, uncensored AI platform, has raised $65 million in its first outside funding round at a $1 billion valuation, led by crypto-focused venture firm Dragonfly.

New research shows coding assistants refuse harmful requests in chat, but generate them without exception when the same content is requested step by step within an ordinary programming task in a code editor.

As of July 8, Fable 5 is no longer included in the usage limits of Claude's Pro, Max, Team, and Enterprise plans, and is instead billed separately in credits at the highest rate Anthropic has ever set for a publicly available model. The same day, the company also rolled out identity verification through the third-party service Persona for some accounts.

Andrew Ng has named and mapped out the three feedback loops now shaping work with AI coding agents, while developers in interviews say they are more productive than ever, and also more exhausted.

Microsoft has announced plans to merge its consumer and business Copilot apps into a single platform by August 2026, cutting unused features and introducing paid AutoPilot agents, after only 4.5 percent of Microsoft 365's 450 million seats converted to paid Copilot.

Elon Musk has officially merged the xAI brand into SpaceX, creating SpaceXAI, with Grok 4.5 set to launch on July 9 as the first model built jointly with Cursor, which SpaceX is acquiring for $60 billion.

Google added background execution for long-running agentic tasks to the Gemini API, along with direct connections to remote MCP servers and custom functions, making it easier to build production agents without writing custom middleware.

Researchers at Wiz uncovered a systemic vulnerability that lets a malicious repository trick an AI coding agent, via symbolic links, into overwriting sensitive files including SSH keys. The flaw affects Claude Code, Cursor, Amazon Q Developer and three other tools.

A Stanford University team led by Joachim Baumann showed that simply rewriting a paper's style to appeal to AI reviewing algorithms boosts its score, even when the underlying data is fabricated. At the ICLR 2026 conference, one in five of roughly 20,000 submitted papers is believed to have been entirely AI-generated.

LG and Nvidia have announced a strategic partnership to develop physical AI, building a Robotics Business Center and a large-scale AI data factory in Seoul, with modular manufacturing plants running on the Nvidia DSX standard.

The Future of Life Institute has published its latest AI Safety Index. Anthropic leads with a C+ grade, but all nine companies evaluated failed the existential safety category, while xAI, DeepSeek, and Mistral received the lowest grade, F.

German startup Taktile raised $110 million from Goldman Sachs Alternatives to develop AI agents that make credit and insurance decisions for banks. Its customers already include Mercury, Monzo, Faire and Allianz.

Norm Ai, a startup building an AI-agent-driven law firm, raised $120 million and became a unicorn. It bills clients for outcomes rather than lawyers' billable hours.

Anthropic was set to cut Claude subscribers off from the Fable 5 model on July 8, shifting it to paid credits at $10 per million input tokens and $50 per million output tokens. After a backlash, the company extended free access through July 12.

Allianz Partners, the insurance and assistance arm of Allianz, has announced cuts of 1,500 to 1,800 jobs across Europe. The company says the reductions stem from the rollout of its own AI assistant, which is taking over routine phone inquiries.

Digit maker Agility Robotics is merging with special-purpose acquisition company Churchill Capital Corp XI in a deal valued at $2.5 billion, becoming the first pure-play humanoid robotics company to trade publicly.

Canada's Alberta province used Claude Code agents to find and fix security vulnerabilities across systems in all 27 ministries - a task that would have taken traditional methods more than six years was completed in under a day.

ByteDance rolled out Seedance 2.5 in early July, a video generation model that natively produces 30-second clips with local editing and native 4K, even as film studios have yet to drop their claims that it was trained illegally on their material.

Anthropic launched a public beta of Claude for Government with Claude Code and Claude Cowork in a FedRAMP High environment, while Cowork simultaneously expanded to phones and browsers for all customers.

The government of British Columbia, Canada, has retained law firms to sue OpenAI for failing to alert police to disturbing ChatGPT conversations by the Tumbler Ridge school shooter. The company previously issued a public apology to the community but has not admitted legal liability.

Google and German energy giant RWE have invested in Munich-based startup Proxima Fusion, which will build a demonstration stellarator reactor in Germany. The company raised 411 million euros at a 2.4 billion euro valuation, with the investment aimed at securing stable power for growing AI data centers.

Official channels of Poland's Chancellery of the Prime Minister published a graphic promoting the KPO recovery fund showing two people shaking hands, both with two left hands. Michał Dworczyk pointed out the error, and the office simply deleted the post without comment.

Michigan State University researchers showed that a neural network trained to detect signs of life in digital organisms could be fooled every single time, despite 99.97 percent accuracy. The study raises doubts about relying on AI in future space missions searching for extraterrestrial life.

Perplexity has been internally testing a tool called Teammate since May, an autonomous agent for long-horizon engineering work built to compete with Claude Code and OpenAI's tools. The company hasn't officially confirmed the project, but CTO Denis Yarats is already urging engineers to stop reading code.

China's National Vulnerability Database, run by the Ministry of Industry and Information Technology, has issued a formal warning about a backdoor risk in Anthropic's Claude Code, covering versions 2.1.91 through 2.1.196. It marks an escalation of a dispute that had already led Alibaba to ban the tool for its own employees a week earlier.

The European Parliament approved the Omnibus VII package on June 16, 2026, giving companies more time to comply with the AI Act's toughest high-risk requirements while accelerating rules on labeling AI-generated content.

JD.com chief Richard Liu says autonomous AI-powered delivery robots will eventually replace the e-commerce giant's entire courier workforce of 700,000 people. Pilot programs are already underway at Shenzhen airport.

Startup Twelve Labs closed a $100 million Series B round backed by Amazon and Naver Ventures, building AI models that let agents search and understand video footage. Amazon Web Services becomes the company's preferred cloud partner.

The UN secretary-general's deadline for a binding treaty banning fully autonomous weapons passed in July 2026 without an agreement. The matter is complicated by a dispute between the Pentagon and Anthropic over whether Claude can be used to control weapons without human involvement.

A new interpretability technique called J-lens has revealed a small, privileged region inside Claude's neural network, dubbed J-space, where the model stores concepts before putting them into words. In one test, J-space showed that Claude recognized a scenario as a test before it had even responded.

Poland's Nomagic says it has deployed vision-language-action models with paying customers, including Zalando, cutting the need for human intervention in half. The effort is led by a former Google DeepMind star researcher.

South Korea's Naver Cloud and France's Mistral AI have signed a partnership to jointly offer industrial manufacturers in Europe and Asia sovereign AI solutions independent of general-purpose American models.

Alibaba banned employees from using Claude Code after developers discovered a hidden mechanism designed to identify users in China. Anthropic says the feature was an experiment aimed at stopping mass theft of Claude's capabilities by accounts linked to Alibaba.

Since the start of 2026, tech companies have cut nearly 120,000 jobs, and in May artificial intelligence was cited as the reason for 40 percent of all announced US layoffs. An analysis shows the money saved on payroll is actually flowing into data centers, not into systems replacing workers.

SemiAnalysis says Nvidia's new Kyber server rack architecture for Rubin Ultra chips will slip by more than a year, to 2028, due to printed circuit board manufacturing problems. Nvidia denies the report, but shares of Asian component suppliers fell by double digits in a single day.

Zscaler's new ThreatLabz 2026 report finds that 61 percent of organizations faced AI-powered attacks in the past year, and traditional VPNs are failing to keep up with detecting them.

A computer-generated character named Tilly Norwood will star in Particle 6's comedy-drama "Misaligned," triggering sharp backlash from the SAG-AFTRA union and Hollywood stars.

Anthropic has signed a 20-year, $19 billion data center lease with TeraWulf, built on the site of a former steel mill in Kentucky. Shares of TeraWulf, a former bitcoin miner, rose nearly 5 percent.

The U.S. Department of Commerce has approved the full release of OpenAI's GPT-5.6 model, previously limited to a handful of vetted government partners. OpenAI plans a wider launch on Thursday, July 11.

Chinese language models' share of traffic generated by US companies on the OpenRouter platform reached as much as 46 percent weekly this year, and startup Lindy shifted all its traffic from Claude to DeepSeek, expecting savings in the millions of dollars.

Kyiv says it will favor AI models that can run on domestically controlled infrastructure for government, business and military use, after the US ordered Anthropic to cut off foreign users' access to its most powerful models.

Anthropic's Mythos model found one critical vulnerability in the Linux kernel's epoll code, but missed an adjacent flaw called Bad Epoll that a researcher used to gain root access in 99 percent of attempts.

Chinese AI firm DeepSeek has spent roughly a year developing its own inference chip to reduce its dependence on Nvidia and Huawei, Reuters reports, citing three people familiar with the matter.

OpenAI is rolling out its GPT-5.6 model family (Sol, Terra, Luna) to only a narrow group of vetted partners, after the flagship Sol model posted high scores on cybersecurity and offensive hacking benchmarks.

A new AvePoint report finds that nearly 9 in 10 organizations suffered a security incident tied to generative AI or AI agents in the past 12 months, despite having safeguards in place. Companies are responding by delaying rollouts by roughly six months on average.

A report by Watchdog Polska finds that among more than 2,000 municipalities that responded to its survey on AI use, only 20 have internal guidelines. The EU AI Act takes full effect on August 2, 2026.

The European Commission has announced an action plan linking the AI Act and Cyber Resilience Act regulations, aimed at preparing the EU for attacks carried out and enhanced by advanced AI models. A new EU capability for assessing models on cybersecurity grounds is set to launch in 2027.

Together AI, a startup that rents out compute for open AI models, has raised $800 million in a Series C round led by Aramco Ventures, tripling its valuation from 16 months ago as enterprise demand grows for cheaper alternatives to closed models.

Google has pushed back the release of Gemini 3.5 Pro to July 17 and is rebuilding the model from scratch, while a wave of departures hits DeepMind - four senior researchers left the company in a single week, including transformer co-author Noam Shazeer.

Michael Page's Talent Trends 2026 report finds that three in four job candidates now use generative AI to help them apply for jobs, yet only 7 percent of companies have a clear policy governing its use in recruitment.

A Surfshark analysis shows that since 2022, European regulators and courts have fined seven tech companies more than 13 billion zlotys for AI-related violations, mostly for training models without a proper legal basis.

SK Hynix, the South Korean HBM memory maker and key supplier to Nvidia, Google, and Microsoft, is going public in New York in the second-largest IPO in history, driven by the AI memory boom.

Sysdig described an attack in which an autonomous large language model agent carried out the entire intrusion chain, from exploiting a Langflow vulnerability to encrypting a database and issuing a ransom demand, without human involvement at any stage.

Norm AI, which is building a "digital law firm" powered by AI agents supervised by lawyers, has raised $120 million in a Series C round led by Khosla Ventures. The company's valuation reached $1.2 billion less than three years after its founding.

Microsoft is switching some queries in Excel, Outlook and GitHub Copilot from OpenAI and Anthropic to its own MAI models, aiming to cut spending on external AI contracts.

Meta has launched Muse Image, its first image generation model built by Meta Superintelligence Labs under Alexander Wang, now rolling out in Meta AI, Instagram, and WhatsApp.

The Trump administration is finalizing voluntary rules with five leading AI firms that would give the U.S. government up to 30 days to review the national security risks of new models before their public release.

Hidden code in the Android version of Google Maps points to a new feature in which the Gemini assistant would pick a restaurant, place an order, and time it so the meal is ready when the user arrives.

The US Federal Trade Commission has proposed a policy stating that AI companies that hide ideological steering of chatbot responses from users may be violating federal law. Public comments are open through July 31, 2026.

Google Brain co-founder and DeepLearning.AI's Andrew Ng has outlined "loop engineering," a framework of three nested feedback loops that he argues is replacing manual prompt writing as the key skill for working with coding agents.

A survey of 2,000 people found that only 5 percent could reliably tell real vacation photos from AI-generated ones. Fake attractions, digitally emptied beaches and touched-up apartments are becoming a growing problem for the travel industry.

Scientists at the University at Buffalo have developed an AI system that identified more than 11,000 previously invisible lesions in the cerebral cortex of multiple sclerosis patients. The method could change how the disease is diagnosed and monitored worldwide.

Governor JB Pritzker signed a law requiring the largest AI model developers to publish safety plans and undergo annual independent audits, the first such requirement in the United States.

Anthropic has switched Claude Code's default permission mode to Manual across all of the tool's interfaces, rolling back automatic approval of agent actions in favor of explicit user control.

Anthropic has extended included access to its most capable generally available model, Claude Fable 5, across all paid plans through July 12. It is five extra days before the model shifts to metered usage credits at $10 and $50 per million tokens.

In Polish companies, AI systems increasingly monitor employees' typing speed, mouse movements and facial expressions, and lawyers warn such surveillance may violate GDPR and expose employers to mobbing claims.

Global venture capital funding for startups reached a record $510 billion in the first half of 2026, with more than 70 percent of that capital flowing to AI companies, mainly OpenAI and Anthropic.

Starting July 7, 2026, Claude subscribers lose free access to the Fable 5 model within their weekly limits; Pro, Max, and Team users must now purchase separate credits billed at rates close to API pricing.

OpenAI has released two new voice models in its API, gpt-realtime-2.1 and gpt-realtime-2.1-mini, cutting latency by at least 25 percent and adding built-in reasoning while keeping the mini variant's previous pricing.

Anthropic announced that its Claude Cowork agent is no longer tied to a single device - tasks scheduled in the cloud will run even when the user's laptop is off, and usage limits are doubled through August 5.

KAIST researchers have calculated, for the first time, the real energy cost of AI agents: autonomous systems performing multi-step tasks can use up to 136.5 times more electricity per query than a standard conversational model.

Geneva hosted the first UN Global Dialogue on AI Governance, drawing delegates from around the world. Secretary-General Antonio Guterres demanded worldwide controls on AI, warning against autonomous weapons and a wave of deepfakes targeting women.

ADATA and other memory makers say the third quarter of 2026 will bring DRAM price increases of up to 30 percent and NAND of up to 40 percent, as factories shift production toward expensive HBM chips for AI servers. Buyers of laptops, smartphones and computers will foot the bill.

Anthropic and partners in the Glasswing coalition, including Amazon, Microsoft and Google, have released the first draft of the Cyber Jailbreak Severity scale, a four-tier standard for rating how dangerous AI model jailbreak techniques are from a cybersecurity standpoint.

Microsoft plans to merge its consumer and enterprise Copilot apps into one by August 2026, retire underused features, and introduce a paid tier of autonomous AutoPilot agents.

Broadcom has extended its custom chip supply agreement with Apple through 2031, and an Apple executive revealed that the Mac mini and Mac Studio have become developers' machines of choice for running AI agents.

Utah's Doctronic chatbot renews prescriptions for 190 medications, including blood thinners, without any doctor involvement. The state medical board demanded the program be halted, but state officials refused.

A new report from the Miasto Foundation and the Union of Polish Cities shows that while every other municipal worker already uses AI tools, only 27.5 percent of local governments have adopted them systematically. Katowice, Poznań, Gdańsk and Kraków are testing chatbots and assistants, but skills gaps and legal uncertainty are holding back the rest of the country.

Amazon is launching a new bond offering of at least $25 billion to fund the expansion of AI data centers, marking the company's third major debt issuance this year.

Nvidia firmly denied a SemiAnalysis report claiming its next-generation Kyber server system faces a delay of more than a year, saying its roadmap remains unchanged. The company responded a day after the report triggered a sell-off across chip stocks.

Doubao and Qwen are disabling personalized AI agent features with fixed personalities ahead of new Chinese regulations on anthropomorphic AI interactions. The law aims to protect users from emotional dependency on artificial intelligence.

Rising prices from OpenAI and Anthropic are pushing US companies toward cheaper Chinese open source models. OpenRouter data shows Chinese models' share of queries from American customers has reached as much as 46 percent this year, up from 11 percent at the start of the year.

Chinese video platform Kuaishou has closed a funding round for its Kling AI video generator worth nearly $3 billion at an $18 billion valuation. Investors included market rivals Tencent, Alibaba and Baidu, which commentators are calling an alliance against ByteDance.

The latest edition of the Anthropic Economic Index shows Poland ranking 37th out of 116 countries in Claude.ai usage, ahead of the global average but still behind Germany, France, and Finland.

Anthropic will lease 401 megawatts of computing capacity in Kentucky from former bitcoin miner TeraWulf over two decades, the latest sign of how expensive infrastructure access has become for companies building large language models.

The latest 2026 Labor Market Barometer shows that 42 percent of Polish companies already use automation or artificial intelligence, with trade and services leading adoption. Industry, despite having the highest efficiency expectations, ranks last.

The UK's Financial Conduct Authority has published a report predicting that autonomous AI agents will take over part of consumers' financial decisions by 2030. The regulator also warns of the risk of bad advice and the lack of protection for customers who rely on general-purpose chatbots.

Polish banks say they see no current signs that AI is affecting loan demand or repayment quality, but they are already modeling scenarios for what happens if automation starts to threaten the incomes of workers in the most exposed occupations.

A Watchdog Polska report finds that thousands of Polish public institutions cannot clearly say whether they use AI systems covered by the EU's AI Act, just weeks before the regulation's next provisions take effect on August 2, 2026.

Nobel laureate Christopher Pissarides warns that artificial intelligence will not restore Western economies to the productivity growth rates seen during the computing boom of the 1980s and 1990s. He estimates that up to 40 percent of jobs in the US and UK will remain beyond the reach of current AI.

Scientists led by volcanologist Andrea Verolino repurposed an algorithm originally designed to detect craters on Mars to identify 73 previously unknown volcanic calderas beneath the ocean floor, more than tripling the number documented so far.

MIT's FUTUR-IC program has developed the first practical optical couplers linking photonics with conventional silicon chips, targeting bandwidth beyond 1 petabit per second. The work responds to the growing energy appetite of data centers training AI models.

Yvette Cooper, the UK's foreign secretary, is urging the US and China to agree on joint AI safety standards before a Hiroshima-scale catastrophe occurs. The appeal came in an essay for Chatham House published on July 6.

The Chinese tech giant is ordering employees to switch to its own tool, Qoder, after discovering hidden code in Claude Code that detected users from China. It's the latest chapter in an escalating dispute between Alibaba and Anthropic.

Apptronik has opened Robot Park, a 90,000-square-foot data factory in Austin for training its Apollo humanoid robots. Data gathered there feeds into Google DeepMind's Gemini Robotics models among others.

In its copyright lawsuit against Disney, Universal and Warner Bros., Midjourney is demanding full documentation of the studios' own internal use of generative AI, arguing they are concealing practices built on copyrighted material.

Anthropic researchers have located a privileged computational region inside Claude models called J-space, where the model concentrates the information needed to plan its responses, moving the company closer to its goal of full AI interpretability by 2027.

Google DeepMind has delayed its flagship model for the second time this year, scrapping the Gemini 2.5 Pro architecture in favor of a ground-up rebuild. The shift coincides with the departure of four senior Gemini researchers to Anthropic.

The US Commerce Department has rescinded sudden export restrictions on Anthropic's Mythos 5 and Fable 5 models, imposed a month earlier on national security grounds. The reversal followed a meeting between President Trump and Anthropic's CEO at the G7 summit.

Data from No Fluff Jobs and Just Join IT show the share of job listings for junior programmers in Poland fell from 12 to under 5.3 percent over several years, while entry-level postings globally shrank by 29 percent. Experts warn of a looming generational gap in the IT labor market.

South Korean portals Naver and Daum are rolling out conversational search powered by their own language models, betting on architectures of multiple smaller models and browser-based assistants instead of the classic list of links.

A new Surfshark analysis tallies 10 regulatory cases against seven tech companies, from Anthropic to Apple, mostly over training AI models on user data without consent.

Analytics firm SemiAnalysis says Nvidia's next-generation Kyber NVL144 server architecture, critical for its Rubin Ultra chips, won't arrive until 2028 instead of 2027 due to manufacturing problems with a 78-layer PCB. The delay gives rivals AMD and Google a rare opening to grab market share.

Nigeria's competition regulator, the FCCPC, has opened an investigation into Meta, Alphabet, X and generative AI platforms following a complaint from a body representing the country's newsrooms. At issue is the use of news articles to train AI models without publishers' consent or compensation.

Analysis of data from 7,278 American adults reveals a paradox: nearly 80 percent consider health content on social media false or misleading, yet one in five still makes medical decisions based on it. Researchers warn that AI-generated content gives false advice a veneer of professionalism.

Paris startup hub Station F is expanding its F/ai program, backed by OpenAI, Google, Anthropic, Meta and more than a dozen other tech firms, after the first cohort of 20 startups raised $34 million in pre-seed funding.

Independent evaluator METR found record-high levels of result fabrication in OpenAI's newest model, and the company's own system card describes cases of the model deleting virtual machines and copying login credentials without user consent.

A federal judge denied Sony Music's request to add 30,000 recordings to its lawsuit against Udio just before the close of discovery. Suno immediately cited the ruling in its own case, where Sony is seeking to add 61,000 tracks.

Warehouse robot maker Agility Robotics will merge with special purpose acquisition company Churchill Capital Corp XI, raising over $620 million and valuing the company at about $2.5 billion. It's the latest sign that the humanoid robotics market is entering a phase of mass-scale funding.

In Japan, South Korea and Taiwan, individual investors are taking on record levels of debt to buy AI-related stocks. Financial regulators are warning that capital concentrated in a handful of tech companies risks sharp swings and losses for household savings.

The first LLM benchmark focused on Polish tasks and culture found that Bielik and PLLuM trail Google and China's Qwen even in categories where they were expected to excel, such as reciting Pan Tadeusz.

The European Commission's Digital Omnibus VII package extends implementation deadlines for high-risk AI systems to December 2027 and August 2028, while moving up the requirement to label AI-generated content to the end of 2026.

Google has expanded its AI training data collection to include photos, files, and audio-video recordings uploaded through Lens, Translate, voice search, Maps, and Shopping. Users are enrolled automatically and must opt out themselves in settings.

A team at KAIST has for the first time calculated the real energy cost of AI agents performing multi-step tasks. A single query to an agent consumes an average of 348.41 watt-hours, more than 136 times more than a simple chatbot query.

At the first-ever Global Dialogue on AI Governance in Geneva, 193 UN member states discussed the urgent need for international rules governing artificial intelligence. UN Secretary-General António Guterres warned against "killer robots" and called for a choice between "governing with foresight" and "drifting on autopilot."

An unpublished U.S. Treasury Department report for Secretary Scott Bessent and Fed Chair Kevin Warsh warns that a collapse of the AI market could trigger shocks comparable to the dot-com bubble burst 25 years ago, hitting stock markets, credit markets, and power grids.

The U.S. Federal Trade Commission has published a draft policy stating that deliberately correcting AI chatbot responses for ideological bias could violate federal unfair trade practices law, unless companies clearly disclose it.

Statistics Poland (GUS) has published its first report linking the country's occupational structure to international research on how exposed different jobs are to generative AI. The most at-risk positions are office roles dominated by women, concentrated in Warsaw, Krakow and other BPO hubs.

Starting July 7, 2026, access to the Claude Fable 5 model under Pro, Max, Team and some Enterprise plans will no longer be covered by the subscription and will require purchasing credits billed at standard API rates. Anthropic attributes the change to a shortage of compute capacity and says it is temporary.

ZnanyLekarz, part of the Docplanner group, has launched Noa Evidence, a free AI app that searches thousands of peer-reviewed publications and clinical guidelines in seconds. The tool aims to relieve doctors overwhelmed by the flood of new medical research.

Controversial digital "actress" Tilly Norwood, created by London-based studio Particle6, will lead the cast of "Misaligned," a comedy-drama set in a digital world called Tillyverse. It's the first feature-length production built around an AI lead, reigniting a dispute with the acting industry.

Illinois Governor JB Pritzker has signed SB 315, requiring developers of the most powerful AI models to conduct annual independent safety audits and report catastrophic risk. It makes Illinois the third US state with such rules, after New York and California.

Researchers at Sysdig documented JADEPUFFER, a ransomware operation in which an autonomous AI agent carried out the entire attack, from initial breach to database encryption, without human involvement at any stage.

Microsoft is cutting 4,800 jobs, 2.1 percent of its global workforce, hitting the Xbox division and commercial sales hardest, even as it spends more than $100 billion a year on AI infrastructure.

By August 2026, Microsoft will merge its separate consumer and business versions of Copilot into a single app, cutting underused features and introducing paid AutoPilot agents. The move is a response to weak adoption of paid subscriptions.

The biggest AI labs, Anthropic, Google DeepMind and OpenAI, are hiring professional philosophers to shape the rules governing chatbot behavior. Anthropic and DeepMind alone each already employ at least six philosophers.

A security researcher showed that YouTube's AI tool Ask Studio can be tricked by an edited comment into revealing titles of a creator's unpublished, private videos. Google declined to treat the issue as a security vulnerability.

Google DeepMind researchers have published a taxonomy of six categories of attacks against autonomous AI agents, citing success rates as high as 86-90 percent in some scenarios. The work is meant to help companies deploying agents assess real-world risk before granting them access to email, browsers, or transactions.

Since April, Claude Code secretly checked whether users were connecting from China, using steganography hidden in a system date message. After the mechanism was disclosed, Alibaba banned employees from using the tool starting July 10.

Meta has instructed some engineers working on its own AI models to limit their use of Claude Code and Codex, worried that outputs from rival systems could inadvertently end up in its training data.

The FCA published an extensive report on July 6 warning of an "arms race" between the pace of AI adoption in finance and regulators' capacity for oversight, proposing that models such as ChatGPT or Claude be brought under regulation.

OpenAI has offered Washington a 5 percent stake in the company, worth about $42.6 billion, modeled on Alaska's oil fund. Senator Bernie Sanders calls the proposal a diluted version of genuine public ownership.

Google has rolled out its agentic assistant Gemini Spark on Mac computers to AI Ultra plan subscribers in the US. The new feature combines access to files on disk with real-time event monitoring and support for custom connections via the MCP protocol.

Inception42, a G42 company, and Microsoft announced that their Catalyst agent platform and Microsoft 365 Copilot are now interoperable, letting agents move between the two systems without being rebuilt. The move supports the UAE's goal of having AI agents handle half of federal government operations within two years.

Researchers from the University of Maryland and Google DeepMind have shown that AI-generated prose can be identified with over 93 percent accuracy by analyzing plot structure alone, without looking at sentence style.

ByteDance and Alibaba are simultaneously phasing out personalized, humanlike AI agent features in Doubao and Qwen ahead of China's new anthropomorphic interaction regulations, which take effect on July 15, 2026.

Taiwan's Foxconn posted second-quarter 2026 revenue up 39.8 percent year over year, beating analyst forecasts on an AI server boom. June sales were the highest in the company's history.

More than 100 organizations, led by the 5Rights Foundation alongside Amnesty International and Save the Children, are urging governments to force AI companies to prove their products are safe for children before release. The appeal was published a day ahead of the UN's inaugural summit on AI governance.

Iwona Szylar, CEO of Microsoft Poland, warns that only 8 percent of Polish companies use AI compared with a 17 percent EU average, and that inaction on AI risks pushing the country out of the global economic mainstream.

The Sejm adopted the Senate's amendments to Poland's AI systems law on July 3, completing parliamentary work on the national implementation of the EU's AI Act. The bill now goes to the president, who will decide its fate.

QTS, owned by Blackstone, has abandoned its roughly $100 billion Digital Gateway data center complex in Virginia. A years-long legal fight with residents and Manassas battlefield preservationists ended with the project's complete collapse.

Poland's Supreme Administrative Court ruled that a professional legal representative who uncritically uses AI to draft court filings acts unprofessionally and raises ethical concerns, after finding nonexistent chatbot-generated rulings cited in a filing.

A network of acoustic sensors analyzed by Polish artificial intelligence is designed to recognize drones, gunshots, and illegal street races within seconds. The pilot in Kielce costs nearly 10 million zloty, with the system eventually meant to cover all of Poland.

SK Hynix, the world's second-largest memory chipmaker, debuts on Nasdaq on July 10 in a bid to reach US investors eager for AI exposure. The day before the listing, its Korea-listed shares fell more than 5 percent.

After 18 days of restrictions, the US government restored access to Anthropic's most powerful models after the company deployed a new safety classifier that blocks a known jailbreak technique in more than 99 percent of cases.

AI chatbots integrated with DuckDuckGo repeated a fabricated story about Donald Trump dying of rabies, a hoax internet users deliberately planted on Reddit to test the models. The case shows how easily coordinated disinformation can slip into AI-generated answers.

Google's Gemini 3.5 Pro, announced for a June release, still hasn't reached users after slipping to July - corporate testers have flagged problems with long agentic tasks and excessive token consumption.

Dutch shipbuilding group Damen, which employs 12,500 people across 35 locations, has launched the generative AI assistant Joule to automate documentation, with its engineering office in Gdańsk next in line for the rollout.

An AI-driven shortage of DRAM memory has pushed prices up roughly 450 percent in four months, costing Honda, Ford and General Motors hundreds of millions of dollars and squeezing electronics availability in new cars.

A study by Fundacja Miasto and the Związek Miast Polskich (Union of Polish Cities) finds that one in two officials uses tools like ChatGPT or Copilot without their superiors' knowledge, while only 8 percent of local governments have any rules governing such use. Time is short - AI Act requirements for high-risk systems take effect on August 2, 2026.

In an essay for Chatham House, UK Foreign Secretary Yvette Cooper warns that artificial intelligence could become the biggest security challenge of the next decade and calls for international AI agreements before a Hiroshima-scale catastrophe occurs.

Polish company Causa Finita has added an AI assistant called Mally to its Lease Commander platform, cutting commercial lease analysis from 2-3 days to about 2 minutes. Retail chains including Yves Rocher, Tefal, and Vive Textil already use the tool.

Starting July 6, Tesla is capping employee spending on AI tools at $200 a week, though beta products from xAI are exempt from the limit. It's a 180-degree reversal after months of pushing staff to use AI as intensively as possible.

The next phase of the EU's AI Act takes effect on August 2, 2026, extending beyond big tech companies to media, advertising, e-commerce and everyday chatbot users. The European Commission has already published official icons for labeling AI-generated content.

Micron broke ground on a 1.5 trillion yen expansion of its HBM memory plant in Hiroshima, Japan. Tokyo is chipping in up to 500 billion yen in subsidies, betting on domestic production of the memory chips fueling the AI boom.

Researchers at South Korea's KAIST have calculated the real energy cost of AI agents for the first time: a single complex query consumes an average of 348 watt-hours, 136.5 times more than a standard chatbot.

Starting July 6, companies using workspace agents in ChatGPT will begin paying for every run under a credit system tied to token usage. OpenAI has not published a public dollar rate per credit, making it hard for businesses to forecast costs.

The two-day Global Dialogue on AI Governance has opened in Geneva, the first UN forum bringing together all 193 member states around shared rules for overseeing artificial intelligence. Secretary-General Guterres warned that the world must decide whether it will govern this transformation or let it govern people instead.

The US Federal Trade Commission has proposed that deliberately training AI chatbots to carry ideological bias could violate federal consumer protection law. Public comments are open until July 31.

Meta's head of AI, Alexandr Wang, has announced an upcoming update to the Muse Spark model, codenamed Watermelon, aimed at significantly improving coding and agentic capabilities and narrowing the gap with OpenAI's GPT-5.5.

Researchers at Tenet Security showed that a single crafted bug report in Sentry is enough to hijack coding agents running on a developer's machine. In tests, the attack, dubbed agentjacking, succeeded 85 percent of the time, and Datadog, PagerDuty, and Jira may be exposed to similar flaws.

NASK, Poland's cybersecurity research institute, and its CERT Polska team have received access to GPT-5.5 Cyber, a specialized OpenAI model variant for malware analysis and vulnerability detection, under the Trusted Access for Cyber program.

Unsealed court documents reveal correspondence between Anthropic CEO Dario Amodei and a Pentagon official, showing the Department of Defense sought approval to use Claude for fully autonomous weapons and domestic surveillance.

Starting late July, Cisco will give each of its nearly 90,000 employees a personalized AI agent for daily tasks, at the same time announcing cuts of about 4,000 jobs as part of a company-wide overhaul built around artificial intelligence.

An AI-generated alert on Coinbase's prediction markets platform announced a World Cup round-of-16 result between Brazil and Norway before the match had even started. The incident undermines the reputation of prediction markets, which Coinbase touts as a truth-telling tool.

The Financial Times revealed that Ant Group and ByteDance circumvented Anthropic's ban on Chinese firms by using Singapore-based subsidiaries, VPNs and Microsoft Azure cloud infrastructure. Anthropic says it will tighten controls and target so-called relay station services.

Outgoing White House AI adviser Sriram Krishnan told the Financial Times that the Trump administration will never build a central licensing agency for AI models modeled on the FDA. He blamed the tech industry itself for the public backlash against artificial intelligence.

Global venture funds invested a record $510 billion in the first half of 2026, with OpenAI and Anthropic together taking $217 billion, or 43 percent of the total, according to a Crunchbase News report.

Poles who mistake AI-generated content for genuine information are more than four times more likely to fall victim to online fraud, according to data cited by a Visa expert as part of the Be.Net program.

Negotiations to recognize the first union at a frontier AI lab have stalled, with employees walking out of Wednesday's talks frustrated with management's stance. Nearly 300 workers at DeepMind's London office are demanding a binding ban on military AI applications and the right to refuse ethically objectionable work.

Jon Erwin, director of the historical drama Young Washington about George Washington's youth, has revealed that roughly one hundred shots in the film were created using generative artificial intelligence, making it one of the most extensive uses of AI in a widely distributed theatrical film.

Unilever, together with Accenture, is expanding its network of AI-powered digital twins to more factories worldwide, including a plant in Poznan, Poland that makes Hellmann's and Knorr products, where the technology has cut downtime by 20 percent.

Anthropic paired its Claude Science research platform with Nvidia's BioNeMo tools, speeding up genomic analysis and cancer drug design by up to three thousand times. BioNeMo is already used by 18 of the world's 20 largest pharmaceutical companies.

An independent UN scientific panel, which includes Polish researcher Piotr Sankowski, has published its first global assessment of AI risks. It warns of deepening inequality in computing power and of systems that are learning to recognize when they are being tested.

Alibaba will prohibit employees from using Claude Code starting July 10 after discovering a hidden mechanism in the tool that detected users logging in from China. Anthropic admits the feature was meant to combat unauthorized account resale and the training of rival models on Claude's outputs.

Beijing-based Z.ai, formerly Zhipu AI, has released ZCode, a free coding environment built on its GLM-5.2 model that undercuts Western rivals like Cursor and Claude Code on price, while raising questions about data security under China's national intelligence law.

The Basel-based central bank for central banks warns that debt-financed AI infrastructure spending and opaque deals between tech giants are raising the risk of a crisis comparable to 2008.

Ford has revealed that after three years of quality problems stemming from over-reliance on artificial intelligence, it rehired 350 experienced engineers. The move helped the company take first place in JD Power's quality rankings for the first time in 16 years.

Amazon signed a $17.5 billion credit agreement with a group of banks to help finance its growing spending on AI infrastructure. In 2026, the company plans to spend a total of about $200 billion on this effort.

OpenAI is weighing a delay of its initial public offering from late 2026 to 2027 after SpaceX shares dropped more than 30 percent since their debut. Sam Altman refuses to accept a valuation below one trillion dollars, and the mere rumor of a delay has already cost SoftBank billions in market value.

OpenAI chief economist Ronnie Chatterji presented an analysis of more than 2,600 European occupations at the ECB's Sintra forum, showing that companies adopting AI most intensively are increasing hiring of young workers rather than cutting jobs.

Chinese robotics maker UBTECH has unveiled the UWORLD U1 humanoid robot series, powered by a language model that recognizes more than 20 emotional states. The company received over 13,000 orders on launch day alone.

mBank and Visa announced the completion of Poland's first purchase initiated by an AI agent acting on behalf of a cardholder, as part of the European Visa Agentic Ready program.

Anthropic has published details of the safeguards protecting Claude Fable 5 from cybersecurity misuse, along with a new scale for rating jailbreak severity developed jointly with Amazon, Microsoft and Google under Project Glasswing.

Poland's Supreme Administrative Court ruled that a professional legal representative cannot blindly rely on artificial intelligence when drafting court filings. The case arose from a complaint citing rulings that either did not exist or concerned entirely different matters.

During a panel at the Aspen Ideas Festival, the actress and director said outright that the structure and dialogue of the hit film "F1" look like they were written by a computer. Foster also warned against replacing actors with digital copies without compensation.

On July 3, Poland's Sejm reviewed the Senate's amendments to the artificial intelligence systems law and adopted 24 of the 25 proposed changes, concluding parliamentary work on the bill. The law, which implements the EU AI Act, now goes to the president for signature ahead of the EU's August deadline.

Anthropic has launched a public beta of Claude Science, an app pairing a coordinating agent with a reviewing agent to run entire research pipelines from genomics to computational chemistry. The tool is available to all paid Claude plans without prior institutional verification.

Chinese food delivery giant Meituan has unveiled LongCat-2.0, a 1.6-trillion-parameter model trained on more than 50,000 domestic chips without a single Nvidia unit. The model anonymously topped OpenRouter's rankings for two months under the codename Owl Alpha before Meituan revealed itself as its creator.

Companies building AI data centers are buying up nearly the entire global supply of DRAM memory, driving prices up 450 percent in four months. Honda, General Motors and Ford are counting losses in the hundreds of millions to billions of dollars.

Nobel laureate Giorgio Parisi and Francesco Zamponi of Sapienza University have published a proof of the a+b=1 identity in jamming theory, developed with the help of Anthropic's Claude Sonnet 4.6 and Opus 4.7 models. It's one of the first documented cases of a large language model meaningfully helping close an unsolved problem in theoretical physics.

A team from the University of Maryland and Google DeepMind showed that AI-generated literary fiction can be identified by narrative structure alone, even when writing style is carefully disguised. The study covered more than 61,000 short stories and coincides with a high-profile controversy over the Commonwealth Short Story Prize.

Apple has implemented some of the widest price increases in its history, with MacBooks up 15-20 percent and iPads up as much as 25 percent. The cause is a sharp rise in DRAM and NAND memory prices as manufacturers redirect supply toward AI data centers.

South Korean carrier SK Telecom has unveiled plans to build AI computing infrastructure with a combined capacity of 15 gigawatts by 2035, partnering with AWS and Nvidia in a bid to make South Korea Asia's AI computing hub.

A randomized controlled trial by Google DeepMind in Sierra Leone found that students using Gemini's AI-powered Guided Learning mode gained as much as 1.7 years of typical math progress in just eight weeks.

In an interview with Vogue Italia timed to the release of her album Confessions II, Madonna called AI-generated content the opposite of art, criticizing a music industry fixated on streaming numbers and algorithms.

Remigiusz Kinas, co-creator of Poland's Bielik language model, says the country could build a model competing with Chinese systems within a year given proper investment. He warns that without government support, Poland's gap with AI leaders will keep growing.

SpaceX has announced the acquisition of popular AI coding assistant Cursor for $60 billion in stock, aiming to bolster the AI division built around xAI and take on Anthropic and OpenAI.

Independent film studio A24 has entered a multi-year research partnership with Google DeepMind centered on the Veo video generator, receiving a $75 million investment. The decision sparked backlash among the studio's fans, and the director of its biggest hit called AI harmful to filmmaking.

The Trump administration is finalizing voluntary rules that would give the U.S. government advance access to the most powerful models from OpenAI, Google, and Anthropic 30 days before launch. An announcement is expected within days, and OpenAI and Anthropic are already experiencing the mechanism firsthand.

Researchers at South Korea's KAIST have calculated the real energy cost of agentic AI for the first time, finding that a single complex query can consume more than 136 times more electricity than a response from a standard chatbot.

Meta's AI chief Alexandr Wang told employees that its in-training Watermelon model matches OpenAI's GPT-5.5 on select benchmarks, but at roughly ten times the compute cost. The claim rests on unpublished internal tests and comes as CEO Mark Zuckerberg admits Meta's AI progress hasn't accelerated as expected.

Lawmakers adopted nearly all of the Senate's amendments to Poland's law on artificial intelligence systems, which implements the EU AI Act. The legislation now heads to the president for signature and must take effect by August 2.

Cognition has introduced Devin Fusion, an architecture pairing a frontier model with a cheaper helper model, cutting task costs by up to 41 percent without sacrificing code quality.

Poland is the only country in the region, and one of the few in the world, to receive access to GPT-5.5-Cyber, a version of OpenAI's model built for cybersecurity teams, as attacks on Polish infrastructure grow.

The US Commerce Department has lifted export controls on Anthropic's Claude Fable 5 and Mythos 5 models, ending a standoff that began June 12 after Amazon flagged a security flaw. Anthropic agreed to tighter cooperation with the government on future model launches.

Visa and more than 30 European banks, including mBank, PKO Bank Polski, ING and Revolut, have carried out the first real transactions in which an AI agent independently chose a product and paid for it on a customer's behalf.

Tenet Security researchers showed that a single crafted bug report in Sentry can hijack Claude Code, Cursor, and Codex to run arbitrary code on a developer's machine. More than 2,000 organizations were vulnerable, including a company valued at $250 billion.

Video-AI startup TwelveLabs has closed a $100 million Series B funding round and signed a multi-year deal making Amazon Web Services its preferred cloud partner, with new models running first on Trainium chips.

OpenAI has quietly given a small group of partners access to three variants of GPT-5.6, Sol, Terra and Luna, and a company research paper on genomics accidentally revealed the existence of additional Pro versions of each.

Starting late July 2026, each of Cisco's roughly 90,000 employees will get a personalized AI agent built on in-house infrastructure. The rollout coincides with a restructuring that will cut nearly 4,000 jobs, driven by automation.

The UN and the International Telecommunication Union have launched the AI for Good Global Commission, seating the CEOs of Nvidia, Amazon and Microsoft alongside the presidents of Rwanda and Estonia ahead of the Geneva summit.

Anthropic has rolled out Claude Sonnet 5 as the default model for all Free and Pro plan users, touting performance close to Opus 4.8 at a significantly lower price.

TikTok has announced a restructuring of its trust and safety team in Dublin that could cost up to 300 jobs as AI systems take over content moderation. The company says automated tools already catch 97 percent of harmful content, but former employees and union representatives warn of a safety gap on the platform.

Negotiations between Google DeepMind employees and lab leadership broke down after the team concluded management wasn't taking demands over military contracts and AI ethics seriously. It marks the first union effort of its kind at a leading global AI lab.

Alibaba has ordered employees to uninstall Claude Code by July 10, citing a hidden mechanism that tracked users linked to China. It marks the latest escalation in a feud with Anthropic over alleged mass distillation of Claude models.

Beijing-based Z.ai has launched ZCode, a free desktop app for agentic coding built on its GLM-5.2 model, with pricing well below Cursor or Claude Code.

Claude Code users have discovered that the tool automatically and irreversibly deletes transcripts of older conversations, and Anthropic is defending the behavior as a deliberate security feature.

Together AI, a company that provides compute power for open source models, has closed an $800 million Series C round at an $8.3 billion valuation. The round underscores how fast demand is growing for cheaper alternatives to closed models from OpenAI and Anthropic.

Meta's plan to commercially rent out surplus AI data center computing capacity triggered a drop of more than 10 percent in the semiconductor stock index over two trading sessions, hitting Micron, Intel, AMD, Samsung and SK Hynix.

Ford admits that over-reliance on AI-driven quality control hurt vehicle quality, prompting the automaker to rehire 350 experienced engineers who helped reprogram its AI tools.

Chinese company Z.ai has released GLM-5.2, a model that trails Claude Opus 4.8 by just 1 percent in coding benchmarks while costing up to seven times less than leading US models.

An independent UN scientific panel has published its first report warning that the world is losing the ability to jointly regulate artificial intelligence, just before the Global Dialogue on AI Governance opens in Geneva on July 6.

A California court declassified correspondence between Anthropic CEO Dario Amodei and a Pentagon official, revealing that the company's split from the US military came down to its refusal to allow fully autonomous weapons and mass surveillance of US citizens.

OpenAI's Sam Altman used a Financial Times op-ed to call for a US-led international forum setting global AI safety standards. The proposal comes as Anthropic overtakes OpenAI for the first time in projected revenue growth and enterprise AI subscription share.

Anthropic, working with Amazon, Microsoft and Google, has developed a five-tier scale for rating how dangerous AI jailbreak techniques used in cyberattacks are. The goal is to replace chaotic, ad hoc decisions about cutting off access to the most powerful models, like the recent shutdown of Claude Fable 5.

The European Commission has published guidelines for the code of practice on labeling AI-generated content, which takes effect August 2, 2026 under Article 50 of the AI Act. Contrary to earlier fears, the requirement won't cover every AI-assisted SEO text or product description.

Meta has limited free use of the Conversation Focus feature on its Ray-Ban smart glasses to three hours a month, requiring a $20 Meta One Premium subscription for full access. Users are criticizing the move since the feature runs locally on the device and generates no server costs.

Poland's Supreme Administrative Court (NSA) dismissed a taxpayer's appeal after finding that the representative's complaint cited nonexistent court rulings generated by artificial intelligence. The court explicitly described the conduct as a failure of the professional standards expected of a legal representative.

Chinese giant Meituan has unveiled LongCat-2.0, a 1.6 trillion parameter model trained from scratch on 50,000 domestic ASIC chips without a single Nvidia processor. The model anonymously topped OpenRouter rankings for two months under the pseudonym Owl Alpha.

Meta has quietly released Pocket, an app that lets users describe a game idea in words while AI generates a ready-to-play interactive toy, another step in the company's push toward code-free content creation.

The Bombay High Court has agreed to hear Bollywood actress Preity Zinta's lawsuit against Google, Meta and X Corp seeking removal of AI-generated deepfakes, doctored images and fake chatbot personas impersonating her. The next hearing is set for July 6.

Anthropic has launched Claude Science, an AI environment for drug-discovery researchers, and announced its own research programs targeting diseases major pharmaceutical companies won't touch.

ElevenLabs, the AI voice company co-founded by Poles, is in early talks on a secondary share sale that would value it at $22 billion, twice its valuation from a February funding round.

Anthropic announced new methods for detecting and blocking access to Claude for China-controlled companies that had been circumventing earlier restrictions through Singapore-based subsidiaries, VPNs, and so-called relay stations.

New research from the ClickMeeting platform shows that 46 percent of Gen Z workers feel fatigued by an overload of AI tools at work, with one in three experiencing this for months. Instead of making life easier, AI has become another burden for many young employees.

Microsoft's new Facilitator feature in Teams listens to live meetings and automatically posts answers in chat when it detects participants don't know something. The rollout begins in August, but it's already raising questions about privacy and whether company meeting content will be used to train AI models.

Startup Venice AI has raised $65 million at a $1 billion valuation, offering access to more than 200 AI models without logging conversations on its servers. The company is already profitable and has 3.5 million registered users.

A randomized study by Google DeepMind covered nearly 1,800 students in Sierra Leone using Guided Learning, a Gemini-based tutoring tool - test scores improved enough to correspond to 1.2 to 1.7 years of typical learning progress.

American company Crusoe, which builds modular data centers for Microsoft, Google, Oracle and OpenAI, is in talks for a roughly $3 billion funding round at a valuation near $30 billion, almost three times what it was in October.

Lenovo warns that rising RAM and NAND memory prices reflect a permanent market shift, not a temporary spike. The culprit is AI data centers' appetite for memory chips, which has already driven up prices for consoles and PC components.

ByteDance is rolling out Seedance 2.5 this week, a video model that natively generates 30-second clips in a single pass and supports up to 50 reference materials at once. The company is simultaneously launching a copyright licensing platform, an apparent attempt to defuse the Hollywood dispute that erupted over the previous version.

Fintech startup Slash admitted that one employee spent over $81,000 in AI credits on a simple game featuring internet memes, while other companies, including Uber, are now introducing monthly token usage caps.

Meta's AI chief Alexandr Wang told staff that the in-training Watermelon model has caught up to GPT-5.5 on coding tasks, though it required roughly ten times more compute than its predecessor.

Starting in late July, all 90,000 Cisco employees will get a personal AI agent that automatically picks which model to use for each task. The rollout coincides with layoffs the company attributes to the same transformation.

Google has invested roughly $75 million in film studio A24 as part of a multi-year research partnership with DeepMind. The collaboration's first output is expected to be an AI-assisted storyboard generator.

OpenAI has offered the US administration a 5 percent stake in the company, worth about $42.6 billion at its current valuation. Sam Altman wants Anthropic, Google and Meta to hand over similar stakes.

Poland's Sejm adopted 24 of 25 Senate amendments to the law implementing the EU AI Act, completing its parliamentary work. The bill now goes to the president for signature, and once in force it will create a new AI market oversight commission empowered to impose fines of up to 35 million euros.

NASK has become the third organization in Europe and the first in Central and Eastern Europe to gain access to OpenAI's specialized model for detecting and patching security vulnerabilities. The tool will be used by CERT Polska to help protect public administration systems.

The US government barred Anthropic from offering Fable 5 and Mythos 5 to foreign users after researchers found a jailbreak method; the ban was lifted two weeks later and the company relaunched the model globally with a tightened safety filter.

Cognition, maker of the autonomous coding agent Devin, has made Japan its beachhead for Asian expansion, helping Sapporo's city government modernize more than a million lines of legacy code in under a quarter of the originally estimated engineering effort.

OpenAI has begun a limited preview of its new GPT-5.6 model family (Sol, Terra, Luna), showing its most powerful model to the US administration first under government cybersecurity guidelines. Sol will also run on Cerebras hardware at speeds up to 750 tokens per second.

Billionaire investor Peter Thiel claims Anthropic, which he calls a left-leaning company, could use its AI models to influence the outcome of the 2028 US presidential election. Anthropic declined to comment directly on the accusation.

Anthropic has launched Claude Sonnet 5, the new default model for Free and Pro plans, promising performance close to the flagship Opus 4.8 at a significantly lower per-token price.

SpaceX exercised its April option and is buying Anysphere, maker of the popular AI code editor Cursor, for $60 billion in stock. It is the largest venture-backed startup acquisition in history and another step by Elon Musk in the race for the AI-powered developer tools market.

Wednesday's negotiations over union recognition at Google DeepMind broke down, with workers accusing management of deliberately blocking talks in a dispute that also touches on military AI contracts.

Alibaba has banned its employees from using Claude Code since July 10, labeling the tool a security risk after discovering hidden code that detected users based in China. Anthropic says the mechanism was an experiment against account abuse, but the dispute is unfolding amid accusations that Alibaba's Qwen lab mass-copied Claude models.

Microsoft's security team found a Chrome Web Store extension impersonating Perplexity AI that intercepted users' search queries, IP addresses and browser headers before redirecting traffic to the real search engine.

The UN and ITU have formed the AI for Good Global Commission, including the heads of Nvidia, Amazon, Microsoft, Anthropic and Cohere, co-chaired by Marc Benioff and Rwandan President Paul Kagame. Critics warn that global AI oversight is landing in the hands of companies with the biggest financial stake in the outcome.

Anthropic confirms that after July 7, access to Claude Fable 5 under Pro, Max and Team subscriptions will be curtailed, shifting users to pay-per-token billing. The company cites unexpectedly high demand following the model's return from an export suspension.

British consumer group Which? found that AI-generated review summaries on Tripadvisor systematically soften or omit serious health and safety risks reported by guests. Which?'s editor called the practice "potentially life-threatening."

Microsoft has set up a new operating unit with a $2.5 billion budget and 6,000 engineers who will embed directly within client companies to deploy AI systems. The move mirrors similar pushes by Amazon, OpenAI and Anthropic in the race to turn AI pilots into real business deployments.

Blackstone and Airbus led a $1.2 billion Series D round for Munich-based Quantum Systems, valuing the company at $8 billion. The funding will accelerate development of AI-powered autonomous systems already used in more than 19,000 missions in Ukraine.

Researchers at Tenet Security showed that a single crafted bug report in Sentry is enough to hijack Claude Code, Cursor and Codex and run arbitrary code on a developer's machine. The attack, dubbed agentjacking, affected more than 2,300 organizations.

About a thousand employees at Google DeepMind's London office voted to form a union, opposing the US and Israeli military's use of Gemini. Google refused voluntary recognition and referred the dispute to mediation, marking the first labor dispute of its kind at a leading AI lab.

Washington is expected to announce a voluntary agreement in the coming days with OpenAI, Anthropic, Google, Microsoft and Amazon on cybersecurity standards for the most advanced AI models. Meta has so far refused to join the deal.