An update on our safety & security practices

Source: OpenAI Date: 2024-09-16 URL: https://openai.com/index/update-on-safety-and-security-practices

Summary

OpenAI’s September 2024 update on safety and security practices — covering the Preparedness Framework’s application to recent model evaluations, red-teaming results, security incident responses, and updated commitments to external safety audits. Published amid increased scrutiny following the departure of several safety-focused researchers in mid-2024.

Implications

Safety/governance thread. The September 2024 safety update arrived in a period of high organizational turbulence — Ilya Sutskever, Jan Leike, and others had recently departed, with public statements about safety culture concerns. The update is partly a trust-restoration document, demonstrating to external stakeholders that safety practices continued despite leadership changes. The Preparedness Framework commitments in the update became the baseline against which subsequent model releases (o1, GPT-5) were evaluated.