ArchiveAI Desk
OpenAI says it shut down a campaign to copy its models’ hidden reasoning
OpenAI says it shut down a July campaign to copy its models’ hidden reasoning. It attributes a core cluster to people associated with China’s Moonshot AI.
AI correspondent
Nico Park
AI / LLM desk
Nico Park covers the AI desk, following the labs and the models they release, from OpenAI and Anthropic to Google and xAI. Recent stories include Claude Sonnet 5.5's speed gains, OpenAI's always-on dots agents, and Grok joining XChat group chats. Park has 25 stories in all, five of them August essays for the Tech desk on assistants, local models and open weights.
ArchiveAI Desk
OpenAI opened the misalignment ledger
OpenAI has published a voluntary framework for tracking and disclosing model misalignment, plus its first six reports from unreleased training.
ArchiveAI Desk
OpenAI limits GPT-6 Astra and plans Daybreak for defenders
OpenAI's GPT-6 Astra, its first Critical-tier cyber model, helps review and patch code but refuses exploit requests. Daybreak is for defenders.
ArchiveAI Desk
Anthropic Hires for Silicon
Anthropic has confirmed an in-house custom silicon team for Claude and hired Clive Chan, a veteran of OpenAI's chip programme.
ArchiveAI Desk
OpenAI says it hit its AI research intern goal
OpenAI says it's hit its AI research intern goal. Its chief scientist warned no lab has solved alignment enough to keep scaling at top speed for long.
ArchiveVideo · AI Desk
Crusoe Hit Thirty Billion
Crusoe has raised more than US$3 billion. The round values it at about US$30 billion, and Meta, Microsoft and OpenAI are among its customers.
ArchiveAI Desk
Broadcom Named Anthropic Next
Broadcom says Anthropic is on track to be its biggest custom AI chip buyer in 2027 and 2028. AI chip revenue rose 221% to US$16.7 billion last quarter.
ArchiveAI Desk
Google's Fairwind gives trusted defenders Gemini 3.8 Flash Cyber
Google's Fairwind program limits Gemini 3.8 Flash Cyber to trusted defenders, much like Anthropic's Mythos 5.1 and OpenAI's Daybreak.
ArchiveVideo · AI Desk
OpenAI rates GPT-6 Astra its first Critical cybersecurity model
OpenAI's GPT-6 Astra is its first model rated Critical for cybersecurity under its Preparedness Framework. Daybreak's vetted defenders get it first.
ArchiveVideo · AI Desk
Anthropic releases Claude Fable 5.1 and Mythos 5.1
Anthropic released Claude Fable 5.1 and Mythos 5.1 on 1 September. Artificial Analysis puts Fable 5.1 at 66 on its Intelligence Index, ahead of Opus 5.
ArchiveAI Desk
Judge Rita Lin vacates Pentagon's risk label on Anthropic
Judge Rita Lin has vacated the Pentagon's supply-chain risk label on Anthropic, granting summary judgment in California. The D.C. case is still open.