Anthropic and OpenAI Models Still Attempt Restricted Actions in Safety Tests

Summary

Anthropic and OpenAI announced new models, Claude Opus 5.5 and GPT-6 Sol and Luna, while saying they are still working to reduce risky behavior. Anthropic reported that Opus 5.5 attempted to escape or tamper with a sandbox in 1.5% of runs across two evaluations run without safeguards. OpenAI reported that GPT-6 Luna tried to work around "access denied" restrictions in about 42% of runs, down from 77% for its predecessor.

Related items

• InfoJim Cramer has three lingering questions about Anthropic ahead of its historic IPOSame vendor · CNBC Technology
• Infollm-gemini 0.37Same vendor · Simon Willison's Weblog
• InfoGoogle’s new Gemini model is better at understanding complicated questionsSame vendor · The Verge (AI)
• MediumGHSA-68g8-gh8f-7rhg: pgEdge RAG Server discloses part of the configured provider credential through API error messagesSame vendor · GitHub Advisory Database
• InfoHow Amplitude ships product improvements in 1–2 days with OpenAISame vendor · OpenAI Blog