Traditional security focuses on systems.
Servers.
Networks.
Applications.
AI agents introduce something new:
Context.
Agents increasingly make decisions based on:
documents
webpages
emails
memory systems
retrieved information
That creates a new attack surface.
An attacker no longer needs to compromise the system.
They only need to influence the information the system trusts.
This is why prompt injection and context poisoning are becoming critical security concerns.
Because the question isn't:
"Can the attacker access the model?"
The question is:
"Can the attacker influence what the model believes?"
This is one of the reasons we're building Crucible.
Pytest for AI agents.
Top comments (0)