Trump Kills AI Security Executive Order After Musk and Zuckerberg Call the President

President Trump cancelled the signing of a long-anticipated AI executive order on May 21, hours before the scheduled White House ceremony. The order would have created a voluntary 90-day framework for pre-launch government review of frontier AI models, with National Security Agency involvement in classified testing. Trump cited concern about hampering U.S. competitive advantage over China, telling reporters he did not want anything in the way of that lead. Last-minute phone calls from Elon Musk, Mark Zuckerberg, and former AI czar David Sacks played a decisive role. The order was originally designed in response to Anthropic’s Mythos model, which demonstrated the ability to identify decade-old security vulnerabilities at scale.

Sources: TechCrunch

Comments (0)

OpenAI Model Autonomously Disproves 80-Year-Old Erdős Geometry Conjecture

OpenAI announced that an internal general-purpose reasoning model autonomously disproved the Erdős unit distance conjecture, an unsolved discrete geometry problem posed by Paul Erdős in 1946. The AI produced a 125-page proof using algebraic number theory, applying Golod-Shafarevich class field towers to connect a geometry question to structures that combinatorial geometers had never previously considered. Fields medalist Tim Gowers called it “a milestone in AI mathematics.” Princeton’s Noga Alon described the result as “outstanding” and “elegant.” Mathematician Will Sawin confirmed configurations now scale as n to the power 1.014. This is the first time AI has autonomously solved a prominent open problem central to an entire mathematical field.

Sources: OpenAI

Comments (0)

GitHub Loses 3,800 Internal Repos to Poisoned VS Code Extension in 18-Minute Attack

GitHub confirmed that threat actor group TeamPCP exfiltrated approximately 3,800 internal repositories after an employee installed a trojanized version of the Nx Console VS Code extension. The poisoned extension was live on the Visual Studio Marketplace for just 18 minutes on May 18, harvesting GitHub tokens, AWS credentials, npm tokens, and 1Password vaults. TeamPCP is asking more than $50,000 for the stolen dataset on underground forums. GitHub says no customer data was affected. The attack is the fourth confirmed wave in TeamPCP’s 2026 campaign, which previously targeted Trivy, LiteLLM, TanStack, and Mistral AI, exposing the AI software supply chain’s vulnerability to rapid automated credential theft.

Sources: VentureBeat

Comments (0)

Anthropic Projects First-Ever Operating Profit at $10.9B Q2 Revenue

Anthropic is on track for its first-ever operating profit in Q2 2026, with revenue projected at $10.9 billion for the quarter, exceeding the company’s total revenue for all of 2025. The figure more than doubles Q1’s $4.8 billion, putting the annual run rate above $30 billion. Expected operating profit is approximately $559 million. Claude Code, which crossed $1 billion in annualized revenue within six months of launch, and broad enterprise adoption of Claude are the key drivers. An IPO is reportedly under consideration for October 2026 at a valuation near $900 billion. Analysts caution that higher scheduled compute costs in the second half of 2026 could pressure margins after June.

Sources: VentureBeat

Comments (0)

Google Gemini Becomes a Full Creative Studio with Adobe, Canva, and CapCut Integrations

Google announced at I/O 2026 that three major creative platforms are integrating directly into the Gemini app: Adobe, Canva, and CapCut. Adobe’s Firefly AI Assistant can now execute multi-step workflows across Photoshop, Premiere, Illustrator, and Lightroom from within a Gemini conversation. Canva’s Magic Layers integration lets users generate an image in Gemini and immediately edit each element on separate layers in Canva; rollout began for AI Ultra subscribers this week. CapCut, downloaded more than 1.2 billion times, will enable video editing and auto-captioning via conversational prompts. The integrations transform Gemini from a standalone generator into a hub that hands AI-created assets directly to professional editing tools without switching applications.

Sources: Google Blog

Comments (0)

Cerebras Runs Trillion-Parameter Kimi K2.6 at 981 Tokens/Sec, Challenging GPU Inference Market

Cerebras Systems announced it is serving Kimi K2.6, a trillion-parameter Mixture-of-Experts model from Beijing-based Moonshot AI, at 981 output tokens per second, independently verified by Artificial Analysis. That speed is 6.7 times faster than the next-fastest GPU cloud provider and 23 times faster than the median. The announcement came days after Cerebras completed the largest tech IPO of 2026, raising $5.5 billion on Nasdaq. Kimi K2.6, released April 20, has become the leading open-weight model for coding and agentic tasks. Cerebras achieves this throughput with custom wafer-scale chips that avoid the memory-bandwidth bottlenecks limiting GPU inference, presenting the most direct challenge yet to Nvidia’s dominance in the AI inference market.

Sources: VentureBeat

Comments (0)

Stellantis and Wayve Target 2028 for AI-Powered Hands-Free Driving Across Vehicle Lineup

Stellantis and London-based autonomous driving startup Wayve announced a strategic partnership to integrate the Wayve AI Driver into Stellantis’s STLA AutoDrive platform, targeting hands-free supervised Level 2++ driving in urban and highway environments by 2028, starting in North America. Wayve’s end-to-end machine learning approach trains on real-world data rather than manually coded rules, enabling generalization across geographies and vehicle types. Engineers brought up a working Stellantis prototype in under two months. The deal expands Wayve’s commercial footprint beyond its London test deployments and builds on Stellantis’s earlier investment in the startup, positioning Wayve as a contender in the race to bring scalable AI driver assistance to mass-market vehicles.

Sources: Stellantis/Wayve

Comments (0)

What's Trending in Tech