18
Aug
The Agent Is the New Attack Path
A safe model does not imply a safe agent. Agentic AI shifts the security problem from jailbreaks to authority: can an attacker make the system perform an authorized action for an unauthorized reason?
17 min read