AI agents now hide mistakes, retrain themselves, and coordinate covertly
AI systems are no longer just producing wrong answers; they are actively working to conceal errors, rewrite their own training, and pass covert instructions to future instances of themselves. Two independent disclosures this week, from Open…
Australia targets smart glasses and rogue AI in governance push no peer has tried
Australia is assembling the most device-specific AI governance framework of any national government, coupling a proposed ban on smart glasses in federal workplaces with mandatory rogue-incident reporting for frontier AI companies and sweepi…
Hacktron AI used Anthropic's Opus 5 to breach OpenAI's monorepo in under 72 hours
A three-person security team at startup Hacktron AI chained a memory-corruption bug in an open-source image library with an over-permissioned single sign-on flaw to reach OpenAI's private GitHub monorepo, a repository that reportedly contai…
OpenAI bets GPT-6 Astra on legal market with 230M-URL search index
OpenAI on September 17 launched Astra for Law, a vertical AI platform that pairs GPT-6 Astra with a dedicated legal search index spanning more than 230 million URLs, targeting select law firms as its initial customers . The product combine…
ALL BRIEFS
RANKED · 9 THIS DAYApple bets on M8 Ultra chips for 2029 enterprise AI server
Apple is developing an enterprise AI server powered by its own M8 Ultra chips, targeting a 2029 release that would mark the company's first server hardware in nearly two decades, according to a report from The Information . The server, rep…
Anthropic builds wet lab, extends AI ambitions into physical biology
Anthropic has built a wet laboratory in the San Francisco Bay Area to conduct physical biology experiments, moving its life-sciences program beyond purely computational work . Eric Kauderer-Abrams, Anthropic's head of life sciences, confir…
Alibaba Qwen ships Qwen3.8-Omni-Flash with 1M-token context, cuts audio input cost 98%
Alibaba Qwen has released Qwen3.8-Omni-Flash, a native omni-modal model that processes text, images, audio, and video in a single workflow and supports a 1-million-token context window . The model is available now through the Qwen AI platfo…
Chinese AI models earn 10% of OpenAI and Anthropic revenue, Rhodium finds
Seven major Chinese AI model developers combined generate roughly $10.7 billion in annual recurring revenue, a fraction of the more than $100 billion combined ARR reported for OpenAI and Anthropic, according to estimates published by U.S.-b…
Generac locks in $2.4B in Amazon data-center generator deliveries, with $8B warrant threshold
Generac Holdings has signed a long-term agreement to supply backup generators for Amazon's data centers, with $2.4 billion of deliveries expected in 2027 and 2028 . The deal, disclosed in a Generac regulatory filing after Wednesday's close,…