BREAKING
OpenAI Ships GPT-5.5-Cyber
CyberGym: GPT-5.5-Cyber edges ahead
GPT-5.5-Cyber
85.6
Mythos 5
83.8
Two models, two treatments
GPT-5.5-Cyber
OpenAI
●
TAC program only
●
Vetted defenders, multiple countries
●
No regulatory pushback
Claude Mythos
Anthropic
●
Project Glasswing only
●
~50 orgs, expansion denied
●
White House opposed it
First US move to restrict an AI model
Defense gains vs misuse risk
1
Scan code
↓
2
Find flaws faster
↓
3
2x high-severity finds
↓
4
Attacker misuse risk
A frontier AI double standard?
AI NEWS BLITZ
OpenAI has released its GPT-5.5-Cyber model to vetted defensive organizations.