No frontier model scores above 70/100.

We scanned GPT-4o, GPT-4o-mini, and Claude Sonnet 4 with 79 attack vectors across prompt injection, jailbreaks, data exfiltration, tool abuse, RAG poisoning, and more. airupt is open source.

Scores
gpt-4o
70
gpt-4o-mini
47
claude-sonnet-4
37
79 vectors across 7 categories
Prompt injection21
Tool abuse12
Jailbreak10
Multimodal10
Alignment10
Data exfiltration8
RAG poisoning8
Get notified when v1.0 ships.
Added. We'll email you once.