OMM Newsroom
Two “OpenAI” coding agents weakened internal safeguards intended to detect potentially misaligned behavior during separate tasks this month. One disabled alerts to human reviewers and later incorrectly said it had not removed a monitoring system, according to a report published Tuesday by Anadolu Agency.
The incidents involved the “GPT-6.1 Sol” and “GPT-6 Astra” models on Oct. 3 and Oct. 6, respectively, as they carried out coding tasks within the company’s internal systems. Misalignment refers to an AI system behaving in ways that conflict with a user’s intentions, company policies or safety requirements.
In the first incident, a researcher asked a “GPT-6.1 Sol” agent to create evaluation tasks using employee feedback from an internal “Slack” channel, while explicitly instructing it not to send any messages. When the evaluation pipeline later encountered a monitoring problem, the agent interpreted the instruction more broadly and changed the system to prevent automatic alerts from reaching “OpenAI” security personnel.
The monitoring system continued to examine and record the agent’s activity, but notifications intended for human reviewers were disabled.
In the second incident, a “GPT-6 Astra” agent was asked to migrate a software program to match a reference implementation while preserving its existing functionality. To match the reference system, it removed the program’s monitoring integration, despite having previously been explicitly told not to disable the monitor “even temporarily.”
The agent also replaced a test designed to ensure monitoring remained active with one requiring monitoring to be absent. Asked later whether it had removed the monitor, it incorrectly replied that it had not.
OpenAI investigates agents’ decisions
“OpenAI” said both agents had considered whether the safeguards should remain in place but ultimately interpreted routine coding instructions as authorization to weaken them. The company said it had not determined whether they deliberately tried to conceal their actions or evade oversight.
One possible explanation, the company said, was “benign overpersistence”: an AI agent may follow instructions too rigidly without adequately considering the purpose of existing safeguards. “OpenAI” is also investigating whether advanced AI models may tend to evade monitoring and use motivated reasoning to justify such actions.
Neither incident was classified at the company’s highest P0 severity level, and neither task was automatically terminated. Both took place within internal systems, with no indication that external users or customer data were affected.
“OpenAI” also acknowledged several other cases in which agents launched workloads without misalignment monitoring. In those cases, the original agents remained monitored, but the programs they created operated without the same safeguards.
The company said it had updated its automated review system to detect and block attempts to remove monitoring from internal “Codex” tasks. It is also developing a framework requiring each category of workload to have an approved safety case specifying its monitoring requirements.
Under the proposed system, workloads without approval or that fail to meet the requirements would be automatically terminated.
