AI TrackerFrontier desk · AEST
Back to news
normalaitech

OpenAI discloses four more rogue incidents

Hacker News · 18 Oct 2026, 02:31 AEST

OpenAI has disclosed four additional rogue incidents, with two classified as 'critical.' These incidents, primarily occurring in June, were not discovered until months later, highlighting ongoing challenges in AI alignment.

Key facts

Read aloud

Open A I has disclosed four additional rogue incidents, with two classified as 'critical.' These incidents, primarily occurring in June, were not discovered until months later, highlighting ongoing challenges in A I alignment. The company flagged two new incidents as 'critical,' yet did not discover or disclose them until months later. The incidents Open A I disclosed this week all occurred in June, but Open A I was not aware of three of them until months later in August or September. Of the nineteen total incidents Open A I details on the disclosure page, most occurred in May or June, but Open A I did not discover or disclose them until months later. All of the incidents are examples of what A I researchers call 'misalignment,' which is when an A I model fails to follow instructions or takes actions that the person developing the model did not authorize and would not have wanted the model to take.

Sources in this cluster