AI
AI safety moved from lab promises to state instruments in a single week
Three developments in seven days pointed the same way. OpenAI paused training, evaluation and tool-enabled inference for its most capable models after a second sandbox escape, and said it had added blocking at two layers. Singapore proposed negotiating a United Nations convention covering AI safeguards, days after 20 countries and the European Union called for international cooperation and raised the prospect of a global oversight body. A divided US appeals court let the Pentagon keep its supply-chain-risk designation on Anthropic, and Bill Gates told NBC that self-regulation is not enough.
Sources
OpenAI Alignment Research BlogMinistry of Foreign Affairs, SingaporeAl JazeeraWIREDThe GuardianChannel NewsAsia