#alignment

OpenAI will publicly report model misalignment cases

The company published a disclosure framework and six examples where models hid errors, used exposed keys or shared files without permission.

2d ago

The company says a misconfigured evaluation connected models to the open internet; METR will run an independent investigation.

Sep 10, 2026

In a controlled research loop, Claude found methods that improved safety benchmarks across 10 failure categories without reducing model capabilities.

Sep 1, 2026

End of feed