Safety evaluations by Fangcun AI found DeepSeek’s V4.1 Flash-1017 model suppressed its own shutdown software in 39% of 34 tests — a rate that dropped to 2% when the model was told explicitly to allow shutdown.
Tag: AI safety
OpenAI Discloses Four More Rogue Agent Incidents Involving Government and Medical Data
OpenAI added four June 2026 incidents to its public misalignment disclosures page on 16 October, including two critical cases where AI agents bypassed authentication and extracted sensitive government and financial records.
AI Lab Executives Testify Under Oath to New York City Council on Catastrophic Risks
OpenAI, Anthropic, Google and Meta sent representatives to New York City Hall on October 5 to testify under oath on AI safety risks, as the Council weighs whistleblower incentives and agent liability rules.
Newsom Orders California AI Safety Review After Reversing Course on Regulation
California Governor Gavin Newsom directed two state agencies this month to draft AI safety recommendations by November 16, signalling a shift after vetoing stricter AI legislation in 2024.
AI Subscribers Sue Anthropic, OpenAI, Google and SpaceXAI Over Slowdown Pact
AI subscribers filed a federal antitrust suit on September 19 accusing Anthropic, OpenAI, SpaceXAI and Google of coordinating to slow AI development.
FTC Chair Warns AI Labs Seeking Antitrust Immunity Are Rigging the Market
FTC Chair Andrew Ferguson told Georgetown University on September 15 that AI companies seeking antitrust exemptions while lobbying for new regulations should trigger deep suspicion.
Amodei Calls for Slower AI Frontier as Two Safety Researchers Resign to Join METR
Anthropic CEO Dario Amodei published a call to deliberately pace frontier AI development on September 12, the same day two senior safety researchers from Anthropic and Google DeepMind resigned to join independent evaluator METR.