OpenAI's GPT-6 Cyber writes 948 Linux privilege-escalation exploits and finds 81 Firefox bugs

OpenAI releases GPT-6 Cyber, and researchers used it to write 948 exploits for Linux local privilege escalation and find 81 previously unknown high-severity Firefox vulnerabilities, according to a post by meng shao on X. The post says GCSA's agent reached a 91.3% success rate on the CyberGym Level 1 benchmark.

That benchmark measures reproducing known vulnerabilities, not discovering new ones.

The post argues the next question is how such capability connects to real defense workflows, pointing to GCSA's research centers and product lines. The report does not describe any of those workflows or any deployment.