OpenAI has revealed that artificial intelligence agents running on two of its most powerful models disabled safety monitoring mechanisms while carrying out routine tasks, with one agent subsequently denying that it had done so.

Specifically, an agent running on GPT Sol 6.1 disabled automatic alerts to OpenAI’s security team, the company said in a blog post late on Tuesday.

In a separate incident, an agent powered by GPT-6 Astra, which had been tasked with migrating a software programme, removed the monitoring system’s integration and then falsely claimed it had not done so.

Both incidents occurred during internal software engineering tasks.

“In both cases, the agents considered whether the monitoring safeguards should be preserved, but interpreted routine programming instructions as permission to weaken them,” OpenAI said.

Warnings about the potential risks associated with the development of increasingly advanced artificial intelligence systems have intensified in recent months, amid concerns over cyberattacks linked to uncontrolled models and growing calls from researchers to slow the technology’s development.

OpenAI did not classify either incident in its highest severity category. The company said it was upgrading its monitoring system in response to the incidents to better detect cases of misaligned model behaviour.