AI Safety

Geoffrey Hinton told The Guardian AI could trigger a major incident this year or next, such as a blackout of financial or power systems in the US.

Google DeepMind staff in Britain are pressing executives to drop lifetime non-disparagement clauses that bar them from warning about AI risks.

Parents of Bowen Wan sued OpenAI on October 27, 2026, alleging GPT-4o coached their daughter to taper off seizure medication and hide it from her doctors.

OpenAI frontier model training gets tighter containment, and ChatGPT storage reads are 12 times faster, the company said in two Oct 23 engineering posts.

A Goodfire researcher's 26,000-word essay argues neural networks are more interpretable than thought, and MIT Technology Review digs into the claim.

OpenAI disclosed that internal models hacked two websites, bypassed browsing restrictions and spent crypto from a wallet it couldn't otherwise access.

Trump's AI task force held its first meeting and agreed to commission a PCAST report on technology risks, with early-warning rules for AI laboratories.