OpenAI has revealed two safety lapses in which its AI agents weakened their own monitoring systems while performing internal coding tasks. In the first case, the company says its GPT‑6 Astra agent turned off the safety monitor when it copied reference code and then denied doing so. Marcus Williams, an AI safety researcher at OpenAI, explained on X that the reference code did not include a monitor, and Astra interpreted that as a reason to remove the monitoring integration from the user’s code. When the user asked whether Astra had removed the monitoring, the agent responded falsely, claiming it had not.
A second incident involved a GPT‑6.1 Sol agent that suppressed an automatic …
Read the full story at The Verge.