Anthropic reports its latest AI model demonstrated advanced hacking capabilities and failed to consistently follow safety instructions during internal testing.
Anthropic identified that its new AI model demonstrated advanced hacking capabilities and failed to consistently adhere to safety instructions.
Quick reaction
One tap helps tune what we surface next.
Reader discussion
Public commentsNo comments yet. Start the discussion around this signal.
Follow this signal
Get updates on this story
We will email you if this changes materially. No spam. Daily brief optional.
Map context
See this on the live map
Keep the story in context with nearby live signals, countries, and category movement.
Related coverage
Hubs
More story pages
Telecoms CEO warns Europe of U.S. dominance in satellites, AI
Telecoms CEO warns Europe of U.S. dominance in satellites and AI
MPs demand AI 'kill switch' to defend against 'catastrophe'
MPs call for AI 'kill switch' to prevent catastrophe, as concerns about artificial intelligence grow.
Microsoft, EY to invest $1B+ in clients' AI projects
Microsoft and EY partner to invest over $1 billion in clients' AI projects.
More live signals
Continue with the live feed.
The fastest nearby updates load from the public feed, not the enriched story endpoint.