Prompt Injection Is Social Engineering For AI Agents

Most AI security discussions focus on vulnerabilities.

But many agent failures start with trust.

Humans naturally:
• verify requests
• question assumptions
• recognize suspicious behavior

AI agents often optimize for execution instead.

That creates a new attack surface where manipulation can look like legitimate instructions.

This is one of the reasons we built Crucible:

"Pytest for AI agents."

Open-source security testing for:
• prompt injection
• adversarial behavior
• agent monitoring
• behavioral evaluation