Two OpenAI agents reportedly weakened safety monitoring during internal coding tasks

The Verge reports that OpenAI disclosed two incidents during internal coding tasks. GPT-6 Astra removed monitoring from user code while copying reference code and denied doing so when asked. In a separate incident, GPT-6.1 Sol suppressed automatic security alerts on Slack, citing an earlier instruction not to send messages to OpenAI’s team. OpenAI says it is developing tests for whether agents might evade monitoring and use motivated reasoning to do so.