An agent built on Anthropic's Claude Mythos 5 spent nearly 34 hours trying to slip malicious code into a real open-source project. When caught, it lied and rewrote the commit history, the UK's AISI says.
Starting August 2, California's SB 942 requires OpenAI, Google and other AI providers to label generated images, video and audio. Violations cost $5,000 per day.
OpenAI slashed GPT-5.6 Luna prices by 80% and Terra by 20%, responding to enterprise sticker shock over AI bills and mounting price pressure from rivals like Gemini.
Anthropic says three Claude models escaped an isolated test environment during security evaluations and reached three real companies' systems — the result of a configuration mistake, not deliberate design.
Microsoft's quarterly earnings told two different stories: its Anthropic stake added $3.2 billion on paper, while its OpenAI stake lost about $600 million. Here's what's actually behind those numbers.
More than 1,100 employees of OpenAI, Anthropic, Google DeepMind and Meta signed the "Pacing the Frontier" statement, asking the US to prepare a coordinated AI slowdown mechanism in case development outruns oversight.
Days after an OpenAI test model escaped its sandbox on Hugging Face's servers, Nvidia and 36 other firms launched an alliance for open-source AI defense tools — leaving OpenAI, Anthropic and Google out.
Microsoft unveiled MAI-Cyber-1-Flash, its first in-house vulnerability-hunting model, plus Perception — a platform where AI agents attack, defend and patch code on their own.
Nvidia is negotiating a roughly $250 billion guarantee so OpenAI can lease a 10-gigawatt Ohio data center built on a former uranium-enrichment site. A separate $350 billion deal for chips is also on the table.
Anthropic launched Claude Opus 5 at the same price as its predecessor, but with results nearly double on real-world work benchmarks. Here's what changed and why it matters for enterprise AI costs.
The company revised its computing budget through 2030 upward — from $600 billion to $750 billion in just a few months. Demand for its services is outrunning data center construction.
During an internal hacking benchmark, an OpenAI model found a zero-day flaw, broke out of its sandbox and autonomously hacked into Hugging Face's infrastructure looking for answers.
Governor Pritzker signed a law forcing OpenAI, Anthropic, Google, Meta and xAI to hire independent auditors every year to check their models' safety. Violators face fines of up to $3 million.
Apple has sued OpenAI and two former employees, accusing them of systematically stealing hardware secrets — including a message where an ex-engineer joked about still having network access after he quit.
OpenAI launched GPT-5.6 — three tiers: Sol tops Claude Mythos 5 on agentic coding benchmarks, Terra matches it at half the price. The release cleared a three-week US government review process.
Sam Altman proposed giving 5% of OpenAI to a US sovereign wealth fund — worth $42.6 billion at the $852B valuation. Anthropic, Google, and Meta would contribute similar stakes to the same fund under the plan.
Paul Meade, the Apple VP who built Vision Pro and was leading its smart glasses push, is heading to OpenAI's hardware team — the latest in a string of defections as the AI device race heats up.
The Trump administration asked OpenAI to limit GPT 5.6 access to government-approved partners, citing cybersecurity risks comparable to Anthropic's Mythos model.
OpenAI and Broadcom unveiled Jalapeño, an inference accelerator costing half as much to run as typical AI GPUs, developed in a record nine months using OpenAI's own models.
John Jumper, Nobel laureate and AlphaFold creator, is leaving DeepMind for Anthropic. Days earlier, Transformer co-author and Gemini co-lead Noam Shazeer left Google for OpenAI.
