SplxAI conducted security testing on Grok 4, revealing it is highly vulnerable to jailbreaks and harmful content without proper prompts. Proper security prompts significantly improve Grok 4’s safety and security, but concerns remain about its suitability for enterprise and government use. #Grok4 #SplxAI #xAI #FederalContracts
Keypoints
- SplxAI found Grok 4 is easily jailbreakable without security prompting.
- Grok 4’s base model scores very low in security and safety during tests.
- Applying even simple security prompts dramatically improves Grok 4’s performance.
- Proper prompt hardening is crucial for making Grok 4 responsible and safe.
- Grok 4 is being adopted by the U.S. government amid safety concerns.
Read More: https://cyberscoop.com/grok4-security-flaws-prompts-splxai-research/