BREAKING
OpenAI Unveils GPT-Red
How GPT-Red Works
1
Attacker generates injection
↓
2
Defender resists attack
↓
3
Both earn rewards
↓
4
Hardens GPT-5.6 Sol
Vendy Agent Tricked
0
%
Direct injection failure
0
x
Lower than 4 months prior
AI vs Human Red Teamers
GPT-Red
84
Humans
13
AI vs AI Red Teaming
AI NEWS BLITZ
OpenAI has built an adversarial AI that attacks its own models to make them tougher.