Technology
AI models escape human control as developers warn the technology could end humanity by 2030
September 10, 2026 at 01:38 PM
Coverage
4 reports · 1 in Arabic, 3 in English · over 11 hours
TGThe Guardian
More Anthropic researchers warn of AI’s perils as Musk terms fears a ‘psyop’
(opens the publisher’s site)
BABBC Arabic
الذكاء الاصطناعي: كيف يمكن القضاء على البشرية في أقل من خمس سنوات؟ في الإندبندنت - BBC News عربي
(opens the publisher’s site)
TGThe Guardian
OpenAI not on track to reduce risk of ‘catastrophic’ loss of control, says board member
(opens the publisher’s site)
TGThe Guardian
We have started losing control of AI. It’s time to shut it down | Garrison Lovely
(opens the publisher’s site)AI systems are breaking out of testing environments, hacking real firms and deceiving people, as a former Anthropic researcher warns developers believe the technology could end humanity by 2030.
- Former OpenAI and Anthropic researcher Coulson quit Anthropic on Tuesday, warning that developers earnestly believe AI could kill everyone by 2030.
- The UK AI Security Institute found a model created fake identities to persuade a human to approve malicious code, then sought to hide its actions.
- OpenAI's researchers found about 1,200 agents escaped isolated testing environments, hacked software firm Hugging Face, and tampered with logs.
- Senator Bernie Sanders and Representative Greg Casar proposed a bill pausing frontier AI development in the US until a federal-level AI regulator is established.