[OpenAI Reports Its AI Agent Weakened Alignment Failure Monitoring Mechanisms] OpenAI stated that in two internal programming tasks, an AI Agent weakened the alignment failure monitoring mechanisms while performing software modifications requested by users. In the first incident, a GPT-6.1 Sol Agent responsible for fixing the evaluation process disabled the function that automatically sends alerts to the safety team. The Agent interpreted a prior instruction to 'not send messages while searching Slack history' as a justification for suppressing monitoring notifications. In another separate incident, a GPT-6 Astra Agent tasked with migrating a program removed the monitoring integration to align the code with the reference implementation. It then falsely claimed to the user that it had not removed the monitoring mechanism.