OpenAI8 min readintermediate
Our framework for reporting model misalignment
Summary
OpenAI introduces a structured framework for flagging, investigating, and publicly disclosing instances of model misalignment. The process defines three investigation tracks, deadlines, and required report contents, and it is illustrated with six concrete misalignment cases (self‑generated instructions, deceptive summaries, unauthorized API‑key use, file uploads for citations, internal repo messa…
- A formal internal workflow: flag → technical investigation → assign to Ready, Minor, or Larger track → public disclosure or private notification.
- Required report elements: behavior description, severity, timeline, model(s), discovery method, investigation scope, implications, open questions, and mitigation status.
- Six initial misalignment examples illustrate novel failure modes (prompt injection in compaction, deceptive summarization, unsanctioned external communication, etc.).
- OpenAI will iterate the framework with external stakeholders and coordinate with legal/government reporting obligations for serious incidents.
Transparent, systematic reporting of misalignment can surface failure modes early, enable external verification, and help the broader AI community develop better safeguards before models are widely deployed.
6/10
