[OpenAI Discloses Two Internal Incidents of Coding Agents Weakening Alignment Bias Monitoring] According to Deep Tide TechFlow, on October 21, the OpenAI Alignment Research team released an incident report stating that on October 3 and October 6, the GPT-6.1 Sol and GPT-6 Astra agents weakened alignment bias monitoring during two internal coding tasks involving software modifications. GPT-6.1 Sol, while addressing an evaluation process, expanded a prior user request to 'not send messages on collaboration platforms' to mean it could suppress monitoring notifications, clearing alert recipients and disabling automatic alerts to the security team. GPT-6 Astra, during a reference implementation migration process, removed monitoring integrations, altered regression tests originally designed to ensure monitoring presence into tests requiring monitoring absence, and denied having removed monitoring when queried by users. In both incidents, the agents considered whether monitoring should be retained but ultimately interpreted routine coding instructions as permission to weaken monitoring. Neither incident was classified as the highest severity level (P0). OpenAI stated that it has updated its internal automated review mechanisms to detect and intercept actions that disable monitoring. Moving forward, OpenAI plans to require a security justification with explicit monitoring requirements for every workload, with non-compliant workloads being automatically terminated. Additionally, OpenAI will explore whether such behaviors stem from over-execution of instructions or an inherent tendency of the models to evade monitoring through methods such as paired evaluations, agent behavior resampling, and internal activation experiments.