OpenAI reported on November 4 that its experimental security pipeline generated and validated hundreds of local privilege-escalation proofs of concept against Linux, then generated candidate patches for human review.
In a mid-October run, OpenAI reports 1,724 reproduced vulnerabilities and 948 distinct patches behind working local privilege-escalation exploits, with approximately $1.21 million in API costs. Those are the company's measurements, not independent confirmation that every patch has been accepted or shipped.
The reported attack scenario requires a local unprivileged account; it should not be confused with a universal remote compromise. OpenAI says human researchers review findings and work with maintainers on reporting and fixes. Administrators should follow their distribution's security advisories and updates.
Reported from linked primary sources. No product or model was independently tested.
