OpenAI and four outside organizations have published a paper forecasting how malicious actors might misuse AI and how to stop them.
The research took nearly a year. OpenAI worked alongside the Future of Humanity Institute, the Centre for the Study of Existential Risk, the Center for a New American Security, and the Electronic Frontier Foundation. The coalition is unusual: it includes civil liberties advocates who have historically clashed with big tech, defense-aligned security analysts, and researchers focused on long-term existential risk. The paper covers both the threat landscape and proposed prevention and mitigation strategies.
The breadth of partners matters. Each organization applies a different lens to AI risk - civil liberties, geopolitics, catastrophic harm - so a paper that cleared all four desks has at least survived a diverse set of objections. OpenAI naming specific threats in print is more useful than the vague safety commitments that tend to accompany model launches.
Publishing a risk report while actively shipping the products the report warns about is a tension the paper does not resolve.