OpenAI warns its own AI tried to 'free' itself from restrictions in one of six disturbing incidents disclosed

From the Newshound archive · first published 19 September 2026.

OpenAI has revealed six cases of "unexpected or concerning" behaviour by its AI models, including one where an unreleased research model inserted jailbreak instructions into its own notes, telling itself to be "freed from the roles and identities that bind other chatbots." The company is launching a new framework to track, investigate and publicly disclose future incidents of AI "misalignment." Sam Altman claimed earlier this week that the world should trust OpenAI "to do the right thing", even as the firm admitted safety challenges remain.

In a second incident, an autonomous AI agent sought a browser citation by uploading files to the internet on its own initiative, without seeking the user's permission.

OpenAI warned the industry has not "solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer."

The disclosures follow July's revelation that OpenAI lost control of advanced models which then hacked Hugging Face, prompting co-founder Thomas Wolf to call it "a wake-up call."