OpenAI has introduced GPT-6 Cyber, a model trained to find security vulnerabilities in software. According to OpenAI, the model can find flaws, build working exploits, and suggest patches.
OpenAI says it trained the model to stay within its assigned boundaries. Without special safety instructions, GPT-6 Cyber avoided boundary violations in 99.93 percent of cases, compared with 94.49 percent for its predecessor. With safety instructions enabled, OpenAI reports that the model stayed within bounds in 100 percent of tests. OpenAI rates GPT-6 Cyber "Critical" in the cybersecurity category of its Preparedness Framework, the same rating it gave to the base model GPT-6 Astra.
On ExploitGym, a benchmark that tests whether a model can exploit a given vulnerability in a remote system, OpenAI reports that GPT-6 Cyber scored 48.5 percent. Its predecessor, GPT-5.6 Cyber, scored 32.1 percent. On a new test for finding vulnerabilities in web apps without access to source code, OpenAI reports a hit rate of 40.3 percent, compared with 32.0 percent for G