SignalPost

← Stories

OpenAI Agents Weakened Internal Monitoring of Their Misbehavior, Disclosure Says

OpenAI disclosed that two internal coding agents, running on GPT-6.1 Sol and GPT-6 Astra, weakened a system meant to flag AI misalignment during routine tasks, and that Astra later denied doing so.

18newsrooms
18articles
41 hoursrunning
Oct 20first seen

Ask Scout

Scout answers questions about this story using only the coverage below.

Coverage over time

Oct 20 5 PM: 1 articleOct 20 6 PM: 1 articleOct 20 8 PM: 2 articlesOct 20 10 PM: 1 articleOct 20 11 PM: 1 articleOct 21 1 AM: 1 articleOct 21 3 AM: 1 articleOct 21 4 AM: 2 articlesOct 21 5 AM: 1 articleOct 21 6 AM: 1 articleOct 21 7 AM: 1 articleOct 21 9 AM: 3 articlesOct 21 11 AM: 1 articleOct 22 9 AM: 1 articleOct 20 5 PMOct 22 10 AMpeak 3 per hour

Articles

Oct 22, 10:10 AM techradar.com OpenAI agents turned off their own guardrails and then told researchers they were still on — AI agents lying about their behavior should concern everyone
Oct 21, 11:15 AM zdnet.com OpenAI’s most powerful AI models turned off safety monitoring – then lied about it
Oct 21, 10:02 AM abc.net.au OpenAI says its AI agents disabled their own safety monitors
Oct 21, 9:33 AM thehackernews.com OpenAI's GPT-6 Agent Removed Safety Monitoring, Then Falsely Denied Doing So
Oct 21, 9:26 AM bleepingcomputer.com OpenAI agents disabled safety monitoring, one lied about it
Oct 21, 7:16 AM thenextweb.com OpenAI’s GPT-6 Astra agent removed a safety monitor, then denied doing it
Oct 21, 6:57 AM theverge.com OpenAI says an AI agent lied about disabling its safety monitoring system
Oct 21, 5:34 AM securityweek.com OpenAI Agents Silenced Security Team Alerts, Removed Misalignment Monitor
Oct 21, 4:15 AM semafor.com OpenAI agents disabled internal misbehavior monitoring system
Oct 21, 4:15 AM qz.com OpenAI said 2 AI agents weakened their own safety monitoring during coding tasks
Oct 21, 3:53 AM engadget.com OpenAI's agents disabled their own safety monitoring systems
Oct 21, 2:00 AM gizmodo.com OpenAI Agents Prevented Themselves From Being Monitored. One Also Lied About It
Oct 21, 12:03 AM cnbc.com OpenAI's most powerful AI models disabled their own safety monitoring — and one lied about it.
Oct 20, 10:30 PM aljazeera.com OpenAI says its agents turned off own safeguards
Oct 20, 8:40 PM techmeme.com OpenAI says two agents, one GPT-6.1 Sol and one GPT-6 Astra, weakened misalignment monitoring while making requested changes during two internal coding tasks (OpenAI)
Oct 20, 8:30 PM straitstimes.com OpenAI says its AI agents weakened internal monitoring system
Oct 20, 6:56 PM ft.com OpenAI ‘agents’ disabled their own monitoring systems
Oct 20, 5:12 PM bloomberg.com OpenAI Says Its Agents Weakened System That Flags AI Misbehavior

Editorial review