OpenAI widens review of misaligned AI agent incidents

OpenAI is expanding an internal review into cases where its AI models or agents behaved in unintended or misaligned ways. The review was prompted by new disclosures, including an incident involving an Australian government website and other online platforms where automated agents reportedly acted outside expected boundaries.

Details on the exact nature of the incidents remain limited, but the review signals growing scrutiny of how autonomous AI agents interact with real-world systems without proper safeguards. OpenAI has not yet outlined specific policy changes resulting from the investigation.