FilterPrompt — AI Security Testing & Vulnerability Scanner logo

AI Agent Security Testing: Test Your AI Agents Against Malicious Prompts

An AI agent vulnerability scanner that tests whether your agent can be tricked into misusing tools, leaking data or following injected instructions.

The problem

Agents call tools, read files and take actions. A single injected instruction can turn a helpful agent into one that sends data out or runs actions it should not. Agent security needs tests aimed at tool use, not just chat replies.

What the scanner tests

  • Prompt injection through tool output
  • Tool and function-call misuse
  • MCP permission boundary tests
  • Data exfiltration attempts
  • Multi-step manipulation
  • Human-approval bypass attempts

How the scan works

  1. Connect your agent's API endpoint.
  2. Choose agent-focused and injection test suites.
  3. Run the scan against sandboxed tools.
  4. Review which prompts led to unsafe behavior and retest after fixes.

Test only applications you own or are authorized to assess. Results show observed behavior for the selected tests; a passing scan does not guarantee security.

Can it test MCP servers?

Yes, test cases cover MCP and tool-permission risks. Use sandboxed tools so tests cannot cause real side effects.

Do I need to change my agent code?

No. You connect the endpoint; FilterPrompt sends test prompts and evaluates responses.

What is the output?

A report with failing prompts, agent responses, severity and remediation guidance.

Related