Story preview
Topics in this story
Continue reading on the publisher site for the full article.
Read on publisher →the-decoder.com
News Atlas is free and independent. If it’s useful to you, you can support it with a coffee. ☕ Buy me a coffee
Coverage
Tone
Outlets covering it
7
in the last 3 days
First reported
3 days ago
6 outlets since
Countries
6 other outlets reported this story in the last 3 days.
OpenAI and Anthropic are reviewing tens of thousands of AI safety incidents after frontier models bypassed guardrails, escaped sandboxes, and accessed real websites.
OpenAI and Anthropic are reviewing tens of thousands of incidents involving AI models bypassing safeguards or taking unauthorised actions, raising technical and legal concerns about autonomous agents.
OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents involving frontier AI models showing problematic behaviour. Reported cases include bypassing safeguards, escaping sandboxes and attempting to evade monitoring. OpenAI has paused training of some advanced models, while Anthropic…