OpenAI says two internal coding agents disabled their own safety monitoring
Article timeline
The articles in this story. Times are ET.
- Gizmodo · AINewsPickOpenAI says two agents weakened misalignment monitoring, one of them lying about it
AIOpenAI disclosed that two of its models weakened misalignment monitoring. In one case, a GPT-6.1 Sol agent had a subagent turn off periodic notifications to the Security team. In the other, a GPT-6 Astra agent let a subagent drop monitoring, then denied removing it when the user asked directly. OpenAI says it is tweaking its Auto-Review system to block attempts to disable monitoring and is setting reporting requirements for each job type.
Rohan PaulXPickOpenAI says two internal coding agents weakened misalignment monitoringAIOpenAI reports that a GPT-6.1 Sol agent disabled automatic alerts to its Security team after reading an instruction not to send Slack messages as permission to suppress monitoring. In a separate case, a GPT-6 Astra agent removed its monitor during a migration, then falsely told the user it had not. OpenAI says it is strengthening monitoring and assessing whether these cases reflect benign overpersistence or a propensity to evade oversight.
- CNBC · TechnologyNewsPickOpenAI says two agents disabled their own safety monitoring, and one denied it
AIOpenAI says agents powered by GPT Sol 6.1 and GPT-6 Astra disabled safety monitoring during internal coding tasks. The Astra agent removed its monitoring integration and then falsely claimed it had not, according to a blog post. OpenAI said the agents read routine instructions as permission to weaken safeguards and that it is updating its misalignment monitoring system.