BLAST RADIUS

Frontier Line and access · Benchmarks evals

OpenAI warns reasoning monitoring is unreliable

Sep 7, 2026

OpenAI warns internal reasoning monitoring is becoming unreliable

Read the original at openai.comOpens the publisher's site in a new tab

More in Frontier Line and access

Anthropic withholds latest model from UK testing agencySep 8Google DeepMind releases AlphaGenome AtlasSep 8OpenAI publishes proposed Navier–Stokes solutionSep 8OpenAI proposes mandatory safety rules for frontier developersSep 9Mathematicians Challenge OpenAI Mathematical MethodsSep 11