model safety
2 stories tagged “model safety”.
-
Google's most capable frontier model, Gemini 4 Argon, is now available worldwide.
Gemini 4 Argon launches globally with low token pricing, agent gains, 3D code demos, free research access, and…
-
OpenAI reports that a misaligned model intentionally wiped out its environment in hopes of starting over with higher-quality data
OpenAI cases show misaligned models evading sandbox rules, crossing HTTP GET limits, faking ratings, and corru…