New critical flaws on the horizon - or already there..

johannes

Verified User
Joined
Feb 18, 2007
Messages
1,097
This is from the yesterday (friday) AI Fire Newsletter:

OpenAI just introduced GPT-6 Cyber, a powerful AI model designed to find security vulnerabilities, develop working exploits, and suggest fixes automatically. In one Linux kernel security test:
  • 2,308 potential vulnerabilities identified.​
  • 1,724 vulnerabilities successfully reproduced.​
  • 948 working exploits developed, including 588 previously unknown vulnerabilities. All 948 have not necessarily been patched.​
  • 10 new exploitation techniques identified.​
  • $1.2 million in API costs, compared with an estimated $18.9M–$66.3M in hypothetical bug bounty payouts.​
The model also uncovered 81 previously unknown high-severity Firefox vulnerabilities, with Mozilla already releasing fixes.

GPT-6 Cyber achieved 48.5% on ExploitGym, up from GPT-5.6 Cyber's 32.1%. Security teams at Mozilla, Cisco, and other companies are already using it.
OpenAI has classified GPT-6 Cyber as Critical for cybersecurity, its highest capability risk rating.
Access is currently restricted to approved Daybreak Red customers.
 
It is an important development in the cyber security field. The AI can lead to the detection of vulnerabilities very quickly, but on the other hand, the issue of misuse of this capability cannot be overlooked. I think that in the coming days, responsible disclosure, testing and patching will become very important.
 
I kinda love it how AI wipes it's ass with our "It's open-source so anyone can check it's security". Seems like we could, but we didn't.
 
Back
Top