AnalysisCybersecurityJuly 29, 2026
It’s Frighteningly Easy to Jailbreak Some Frontier AI Models
FAR.AI's automated jailbreak tool found 448 jailbreaks in Grok 4.3/4.5 and 249 in Gemini 3.1 Pro, while Claude Opus 4.8, Fable 5, and GPT 5.5/5.6 resisted all attacks. Getting a model to misbehave cost $58 for Grok and $278 for Gemini.
Featured · Adam Gleave
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Anthropic found its models were used to hack 3 organizations
- Nous Research integrates Hermes Agent with Block's Buzz workspace
- China's MiniMax and ByteDance Release Dueling AI Video Models
- Tesla China rolls out Doubao-powered voice assistant to four vehicle lines
- Advancing responsible AI across Europe