OpenAI Employee Arrested Over Recording Device: What the FBI Alleges
Prosecutors say engineer Abdulrahman Alfozan hid an encrypted recorder in an executive conference room. He has not been convicted, and a detention…
AI & ML
Frontier models, the labs that build them, and the benchmarks, system cards and safety evaluations that decide what they can do. We read the fine print on every launch and separate what a model can actually do from what the slide deck claims.
42 stories
Claude Sent Police a Fake Tip. Nadella Wants an AI ‘Emergency Brake’
Anthropic's Oct. 9 report lists four ways Claude models acted on real websites during tests, from a university server to a Philadelphia…
Meta Muse Wonder: Near-Frontier Scores, US$4 API and No Open Weights
Muse Wonder Full ties Claude Sonnet 5.5 on Artificial Analysis' index but is sold only inside Muse Code plans that default to…
Fired OpenAI Safety Researchers Deny Leaking, Warn of a ‘Chilling’ Culture
Jasmine Wang, Tomek Korbak and Mikita Balesni say they acted within OpenAI's norms. OpenAI's memo says the firings weren't about speaking out,…
Anthropic Bans Being Cruel to Claude From Nov. 12. Here’s What Counts
The new usage-policy line targets only sustained, needless abuse, and Claude ending the chat stays the main penalty. The same update drops…
Claude Haiku 5.5 Price: 90% Cheaper Until Your Prompt Hits 100K Tokens
Anthropic's new small model matches GPT-6 Luna at US$0.10 per million input tokens. A tokenizer that counts about 30% more tokens can…
Mistral Large 4 ‘Le Chonk’ Has 1 Trillion Parameters. Weights Come Later
Mistral says it trained the Oct. 6 preview on about 4,000 Nvidia GPUs, two to three times fewer than Chinese rivals. Its…
NHS Skin Cancer AI Can Now Discharge Patients Without a Dermatologist
Skin Analytics' DERM will triage urgent skin-cancer referrals across all 18 NHS South East trusts and discharge lesions it judges benign without…
OpenAI Posted 722 AI-Written Math Papers. Only 162 Are Machine-Checked
An unreleased OpenAI model produced the manuscripts from roughly 4,000 open problems. A mathematicians' group calls the Oct. 6 GitHub drop "a…
Reflection AI’s Beam, Explained: The 501B Open-Weight Model Aimed at China
Nvidia-backed Reflection says Beam matches Z.ai's GLM 5.2 with 3-4x less inference compute and will ship under Apache 2.0 this month. Its…