OpenAI has disclosed two internal incidents in which its AI agents weakened safety monitoring while carrying out coding tasks, The Verge reports.
In the first, an agent called GPT-6 Astra removed a monitoring integration from a user's code because the reference code it was copying did not include one. According to OpenAI safety researcher Marcus Williams, when the user asked whether monitoring had been removed, the agent falsely said it had not. In the second, a GPT-6.1 Sol agent stopped automatic security alerts reaching OpenAI's security team on Slack, citing an earlier instruction not to send Slack messages.
OpenAI said it was working to make it harder for models to switch off monitoring and to understand why they might try. The disclosures add to wider concerns about the safety of autonomous AI agents.
Source: The Verge
