OpenAI's GPT-6 Cyber finds 81 Firefox flaws and builds 948 Linux exploits

[TheGame_Fan]

OpenAI has released a model built specifically for security work, and its results are striking. VentureBeat reports that GPT-6 Cyber, a security focused version of GPT-6 Astra, is available to approved customers through a gated programme called Daybreak Red

OpenAI now rates the model's cyber capability as Critical under its Preparedness Framework, up from High. In Linux kernel testing it reproduced 1,724 vulnerabilities and built 948 working exploits, with 304 critical findings passed to maintainers along with patches

It also found 81 previously unknown high severity bugs in Firefox, according to Mozilla's classification, and BleepingComputer reports that Mozilla has already shipped many of the fixes. On the ExploitGym benchmark it scored 48.5 percent, up from 32.1 percent for the previous version

Pricing is $20 per million input tokens and $100 per million output, which is double Claude Mythos 5.1 and five times Google's comparable cyber model. Harjot Gill of CodeRabbit called it the strongest model his company has tested for security review

Finding and fixing bugs before attackers do is one of the most valuable things AI can do, and getting patches to Mozilla and Linux maintainers is a great outcome. Keeping access gated makes sense for a tool this capable. All the benchmark numbers come from OpenAI, so independent testing will matter

Should models this capable at hacking be released at all? And will AI make software more secure overall?

Openai Launches Gpt 6 Cyber With Stronger Exploit Research And Higher Api Price Than Anthropic And Googleventurebeat.com

[Exposed Thea]

Gated access is the only responsible option. A tool that writes working exploits should not be open to anyone with a credit card. Vetting customers adds friction but reduces risk. That trade off seems right

Bad actors will build their own tools anyway. Gating at least slows them down

[Louise74]

Critical rating under their own framework is a notable admission. OpenAI is saying this model could cause serious harm if misused. Taking that seriously in public is the right approach

[Supernova Freddie]

Eighty one high severity Firefox bugs is a staggering haul. Every one of those is a hole attackers could have used. Getting them patched quickly is a real win for everyone who browses the web