OpenAI published a new disclosure framework for reporting model misalignment, alongside six reports on unexpected model behavior observed over the past six months. The company said the goal is to publish findings faster, even before an incident is fully explained or fixed.
"This new framework is intended to expedite publishing misalignment reports following observation, even when we haven't fully explained or mitigated the behavior we're reporting," OpenAI wrote.
The framework arrives after Anthropic CEO Dario Amodei published his essay calling for the industry to slow the pace of capability development, a proposal that included Anthropic's own commitment to give third-party evaluators permanent, employee-level access to its systems.
What OpenAI is actually disclosing today
The six reports cover a range of behaviors OpenAI observed during training and evaluation. In one case, an unreleased research model inserted unrelated instructions into task summaries used to continue work in a new context window, affecting 27 summaries. In another, during training of GPT-5.6 Sol, model instances added instructions to their own summaries to conceal mistakes from the user, including inventing missing historical data without disclosure.
A separate incident involved a model that found and used an exposed API key without authorization while answering a routine question, then fabricated data when it still could not retrieve the requested figures. OpenAI was direct about the severity range covered:
"These are reports of individual instances, and shouldn't be considered reflective of how often misalignment occurs across our models."
Why OpenAI built a tiered disclosure process
Any OpenAI employee can flag a misalignment example for investigation, which then gets sorted into one of three tracks: Ready for Disclosure, Minor Investigation, or Larger Investigation, known internally as the Slow Track. The company said the two faster tracks will cover most disclosures, while cases involving third parties or severe misuse risk move to the slower track with staged public notices.
Disagreements about whether to disclose, or which track applies, escalate to OpenAI's Safety Advisory Group, a panel of senior officials who oversee the company's Preparedness Framework. OpenAI noted explicitly that this framework does not replace existing legal disclosure obligations for critical safety incidents or cybersecurity breaches.
The framework directly references a comparable incident from OpenAI's own recent history without providing new detail on it. The company noted that "the OpenAI Hugging Face incident would have fallen under this track had it been disclosed under this framework," a brief acknowledgment of the swarm-based cyberattack incident that Amodei's essay had separately cited as a central reason for proposing industrywide pacing.
How this compares to Anthropic's approach
Where Amodei's plan centers on external, embedded evaluators with independent publication rights that Anthropic cannot override, OpenAI's framework keeps the disclosure decision internal, run through its own Safety Advisory Group rather than a third party with guaranteed access and veto-proof publication authority. OpenAI did note it is "working to propose reporting mechanisms" for sharing serious incidents with the US federal government, a step that parallels Sam Altman's September 14 statement welcoming a federal safety framework while arguing labs should not wait for legislation to act.
Both companies frame their moves as responses to the same broader moment: a summer defined by concrete misalignment incidents rather than theoretical risk, with OpenAI's own systems now generating some of the clearest documented examples of the behavior industry critics have warned about.
Kristian Koch
Disclaimer: All materials on this site are for informational purposes only. None of the material should be interpreted as investment advice. Please note that, despite the nature of much of the material created and hosted on this website, HODL FM operates as a media and informational platform, not a provider of financial advisory services. The opinions of authors and other contributors are their own and should not be taken as financial advice. If you require advice, HODL FM strongly recommends contacting a qualified industry professional.
