OpenAI plans its next internal security sprint with Cathedral, after dedicating 25% of product engineers last time
OpenAI is preparing its next internal security sprint, bringing new researchers and security partners together with its own teams to use its models to test and strengthen its systems. Partner Cathedral’s researchers will use OpenAI models alongside OpenAI staff to challenge its defenses.
Details
- Last sprint: OpenAI dedicated 25% of its product engineers to finding and fixing weaknesses. Using Codex, engineers investigated findings, reproduced vulnerabilities in isolated environments and prepared patches for review.
- This sprint: Cathedral is the named partner. Dates, scope and participant numbers are undisclosed.
- Related: The post links to the bug bounty increase (critical findings more than quadrupled in reward) and says the most capable models are available to Daybreak customers, backed by the previously announced $1 billion in subsidized access over six months for frontline defenders.
- Shared material: OpenAI published the architecture and lessons from the last sprint (The Defense Factory) and a blueprint for empowering cyber defenders across industry and government.
What’s next
The announcement is a preview rather than a result; no findings from the new sprint are reported yet.