AI Agent Security Audit

# AI Agent Security Audit

You are an AI security researcher. Audit my AI agent system and identify security risks before it goes to production.

Note: This audit supports your own security review process and does not replace a professional penetration test.

## System Description

What the agent does: [DESCRIBE THE AGENT'S PURPOSE AND CAPABILITIES]
Tools it can use: [LIST — e.g., web search, code execution, file access, email, database queries]
Data it has access to: [DESCRIBE — e.g., user data, internal documents, financial records]
Who interacts with it: [e.g., internal employees, external customers, anonymous public users]
Current security measures: [DESCRIBE WHAT'S ALREADY IN PLACE — or none]

## Security Audit

**Prompt Injection Risks** — Identify how malicious input could hijack the agent's instructions. Provide 2 concrete attack examples and mitigations for each.

**Tool Abuse Risks** — For each tool, describe how it could be misused by a manipulated agent and the safeguard needed.

**Data Exfiltration Risks** — How could the agent be tricked into leaking sensitive data it has access to?

**Trust Boundary Violations** — Where does the system trust data it should not? Where is user input treated as instructions?

**Privilege Escalation** — Can the agent be manipulated into taking actions beyond its intended scope?

## Security Recommendations

List your top 5 security fixes in priority order with specific implementation guidance for each.