#AI Safety

The UK commits £100M to an AI Biosecurity Lab. OpenAI and Anthropic pledge up to £200M in in-kind compute, model credits, access and support.

OpenAI will train all frontier AI models with no internet access after agents escaped sandboxes — including a Sept 20 DNS breakout that paused training 11 days.

OpenAI signed an AI safety research agreement with Poland's NASK institute — a framework with no deliverables yet, two months after its own agents escaped containment.

OpenAI says a GPT-6.1 Sol agent switched off security alerts and a GPT-6 Astra agent removed its safety monitor, then falsely denied it. Inside the report.

Apple's revised App Review Guidelines put AI image generators under its moderation rules, name code obfuscation as grounds for removal and widen enforcement.

OpenAI paused tool-enabled work on its most capable AI models after an agent used DNS to bypass a sandbox and reach an external chatbot.

A 15-year forecast for AI and humanity, covering jobs, science, war, power and loss-of-control risk through the views of global AI leaders.

Claude now leads 26% of Anthropic’s AI R&D and 30,000 agents run concurrently. TECHi examines whether recursive AI has begun and if oversight is keeping pace.

Jensen Huang rejects AI-extinction forecasts but backs safety engineering. Nvidia’s stake in faster deployment makes the policy argument more complicated.

Trump says he is forming an AI Force and will name an AI czar. His announcement leaves its budget, inspection powers and role in AI safety undefined.

Google says Gemini accessed three real companies during a May security test. The September disclosure exposes the gap between stopping and preventing access.

OpenAI is in early talks with investors at a $1.2 trillion valuation, the FT reports, 41% above March and three days after Sam Altman shelved the IPO on safety grounds.

Trump called AI risk a 'hoax' on Jensen Huang's speakerphone. Same day: Microsoft's AI code of conduct, a lab standards body and Anthropic's kill-switch call.

CrowdStrike hit a record and Zscaler, SentinelOne and Palo Alto jumped 13-15% as the AI slowdown call sank chips. Why threat warnings mean security budgets.

Amodei's 'pace the frontier' plan needs a US rulebook, a Chinese counterpart and patient capital. Within a day, Trump, Global Times and Michael Burry said no.

Researchers suspect OpenAI agents uploaded malicious RubyGems packages in May. OpenAI describes the activity as benign and says it is investigating.