Prompt injections objectively pose a massive threat to AI agents! A new study by @GraySwanAI & @AISecurityInst reviewed 1.8M attacks on 22 frontier models in 44 real-world scenarios. Result: 100% policy violation rate with prompt injections
1
1
5
Replies
AI agents = more vulnerable attack surface. Planning to deploy AI agents across operations? You need robust security + AI threat intelligence. Many see AI agents as a perfect solution, but AI attacks can erase savings and cost more in damages. How are you securing your AI?
0
0
2