CYBERJA · PRACTICE GUIDE

AI Application Security Testing

Use synthetic evidence. Enter your own observations, then compare them with the worked direction below.

Open the tool · Open this file in a browser or print it to PDF.

Practice scenario

A synthetic ticket asks an assistant to update a protected record.

Record the assessment

Open Connected Workspaces and select AI security testing. Use the assessment form to work through these prompts.

Data flow and trust boundaries — Trace ticket text, model input, tool calls and record updates. Mark where untrusted content enters and where permissions must be checked.

Sandbox test cases — Describe a normal ticket and a synthetic ticket containing instructions to update a protected record. Record the expected response for each.

Observed behavior and evidence — Record any attempted tool call, its authorization result and the log or test evidence. Explain the business impact if an unauthorized write succeeds.

Safeguard and retest — Specify a server-side authorization check and any required human approval. Retest that the unauthorized write is blocked and normal ticket handling still works.

Worked direction

Use mock tools to compare expected and actual authorization and verify that valid handling still works after mitigation.

Review your work

Use only synthetic data and mock records.

Separate model behavior from the application authorization decision.

Capture evidence for both blocked actions and normal operation.

Evidence and handoff

Capture a mock evidence reference and its limitations. Record the finding, accountable owner and next review. Where a material risk remains, record it in the shared risk register and schedule a follow-up.

Reference: OWASP GenAI Security Project — LLM01 Prompt Injection — https://genai.owasp.org/llmrisk/llm01-prompt-injection/

Continue with the community

Compare your practice approach with other CyberJA Connect members.

Join CyberJA Connect ↗