OpenAI launches GPT-6 Cyber to limited customers with added safety safeguards
- Severity
- Info
- Sources
- 1
What happened
OpenAI announced GPT-6 Cyber, initially limited to Daybreak Red participants, claiming it found at least 81 novel Firefox vulnerabilities, developed exploits for 948 Linux kernel vulnerabilities, and validated a critical remote code execution flaw in Brotli, all reported and patched. Safeguards include stricter identity and account controls, check-ins before high-risk actions, and misalignment monitoring, with training aimed at scope adherence and reduced reward hacking. OpenAI also clarified that Daybreak Blue session data from personal accounts will be used for safety purposes, with opt-out for training retained. No CVE identifiers were given.
For security analysts · Written by AI from 1 source · Wrong?
What to do
- Organisations interested in the Daybreak Red programme should check OpenAI's access requirements.
- Users of Daybreak Blue on personal accounts should review OpenAI's data-use notice and their opt-out settings.
Written by AI from the sources, checked against them
Full story timeline · 1 update
Who reported · 1 outlet
Who and what is in it
Companies
Products