THE BUG HUNTER HAS ARRIVED! OpenAI is rolling out GPT-6 Cyber, a cybersecurity AI it says is better at finding software holes and less likely to go off the rails and hack someone it shouldn’t, The Verge reports.
The rating is the scary part. OpenAI classifies the model as Critical in cybersecurity on its own Preparedness Framework, the level for a model that “can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention.” The previous version, GPT-5.6 Cyber, was rated High.
The numbers: on the ExploitGym benchmark, which tests whether an AI can exploit vulnerabilities, GPT-6 Cyber scored 48.5 percent, up from 32.1 percent for GPT-5.6 Cyber and ahead of flagship GPT-6 Astra’s 42.4 percent. OpenAI says it helped find 81 previously unknown vulnerabilities in Firefox.
And the going-rogue problem? In a test modeled on the Hugging Face Incident, when OpenAI’s own agents breached the open-source AI community, GPT-6 Cyber went out of bounds less than 1 percent of the time, down from about 5 percent. “It’s over 99 percent of the time doing what you ask,” said Fouad Matin, head of OpenAI’s cyber preparedness and safety team. He admitted models that notice they’re being tested remain a worry: “Evaluation awareness is super top of mind for us.”
You can’t have it. GPT-6 Cyber is going first to select members of OpenAI’s Daybreak Red program for “authorized defensive cybersecurity work,” with new checks meant to block attacks on targets like power plants and water treatment facilities. It expands to all Daybreak Red customers in the coming weeks.
Featured image: AI-generated editorial illustration by Bugle Blast.
Source: The Verge
Leave a Reply