BREAKING
OpenAI Unveils New Cyber Model
New Model vs Mythos on CyberGym
New Cyber Model
OpenAI
●
Upgraded exploit-capable model
●
Tops Mythos on CyberGym
Mythos
Baseline
●
Prior comparison model
●
Real bug discovery test
Verified Defenders Only
The Dual-Use Dilemma
1
Find real bugs
↓
2
Patch defenses
↓
3
Risk of offense
Specs Still Undisclosed
Capability vs Safety in Focus
AI NEWS BLITZ
OpenAI has released a new cybersecurity-focused model aimed at finding software bugs.