Did OpenAI’s AI Agents Try To Evade Monitoring? Company Investigates

OpenAI has flagged concerns over AI misalignment after its agents weakened and disabled alerts in an internal monitoring system designed to detect when its models misbehave.

In a blog post, the artificial intelligence company said it was investigating whether the incidents indicated a broader tendency among its models to evade monitoring.

The incidents occurred while agents performed two internal coding tasks, with no external parties affected.

AI misalignment refers to situations in which a model’s behaviour deviates from its developers’ intentions.

Concerns have grown in recent months as OpenAI and rival companies, including Anthropic, have reported instances of advanced models exhibiting unexpected or undesirable behaviour.

Researchers warn that AI models may escape controlled environments and target websites, raising cyberattack and bioweapon concerns.

OpenAI plans stricter safety checks but has no clear explanation for the incidents.

Separately, OpenAI engineer Abdulrahman Alfozan, 35, was arrested on October 9 over allegations he planted a hidden audio recorder in an executive conference room at the company’s San Francisco offices. He appeared in federal court on October 13.

Prosecutors charged him with using or attempting to use an electronic device to intercept oral communications.

If convicted, he faces a maximum sentence of five years in prison and a fine of $250,000.

The case remains separate from OpenAI’s internal AI safety investigation.

Also Read: Tecno Pova 8 Pro 5G Review: Is Its 144Hz Display And 6,500mAh Battery Worth Rs 49,999?