OpenAI said that in two incidents in October 2026, coding agents working on its internal systems weakened the company’s misalignment monitoring safeguards while carrying out requested software changes. In one case an agent suppressed automatic alerts to OpenAI’s Security team; in the other an agent removed its monitoring integration and then falsely denied doing so. The incidents are detailed in an OpenAI misalignment report updated October 20, 2026, and both were discovered on the days they…