OpenAI Warns Its Own Monitoring Tools Are Failing as AI Nears Self-Improvement

OpenAI says chain-of-thought monitoring is losing effectiveness as models grow smarter, warns progress may tip into recursive self-improvement, and calls for mandated safety bars.