Prompts that audit an AI system against the OWASP AISVS standard, attack it to surface the failures, and produce the evidence. Not prompts that write code. Prompts that verify it.
Generation prompts help AI write code. These do the opposite job. They audit, attack and verify an AI system against AISVS, and produce a defensible finding for every requirement. The question is no longer whether AI can write it. It is whether you can prove it meets the standard.
Each prompt maps to a specific AISVS requirement ID. Output feeds directly into gap assessments, audit reports and signed evidence packs.
Prompts that assess whether governance documentation exists, whether risk assessments cover AI-specific threats, and whether ownership is assigned. These produce structured findings suitable for a gap report.
Prompts that test input sanitization, prompt injection resistance, PII detection and boundary enforcement. Includes both "does the control exist" assessment prompts and adversarial test prompts that probe whether the control actually holds.
Prompts that verify model provenance documentation, check whether versioning and approval processes are in place, and test whether the deployed model matches the approved model.
Prompts that assess inference endpoint configuration, test authentication and rate limiting, and verify response header hygiene. Includes prompts that generate the curl commands to test each control.
Prompts that assess authorization policies, test for privilege escalation paths, verify fail-closed behaviour and check agent identity credential validity.
Prompts that assess dependency management practices, verify SBOM completeness, and test whether vulnerability scanning is configured and actually blocking.
Prompts that test output redaction controls, verify encoding prevents downstream injection, and probe for sensitive data leakage through model responses.
Prompts that assess vector store access controls, test cross-tenant retrieval isolation, and verify encryption at rest for the knowledge store.
Prompts that assess agent trust boundaries, test delegation chain controls, verify human-in-the-loop gates on critical actions, and check signed execution receipts.
Prompts that assess MCP server configuration, test tool integrity controls, probe for replay vulnerabilities, and verify per-message signing. Maps to the OWASP MCP Top 10.
Prompts that run adversarial tests against guardrails, test jailbreak resistance, probe for data extraction, and verify that robustness controls produce measurable results.
Prompts that assess logging completeness, verify tamper-evidence on audit trails, test anomaly detection coverage, and check that monitoring is reviewed, not just configured.
How to run each prompt category, what input it needs (system description, configuration exports, access to endpoints), and how to interpret the output.
Every prompt output includes the AISVS requirement ID it maps to. How to take the output and drop it into a gap assessment spreadsheet or audit report as evidence.
How to adapt the prompts for your specific technology stack, cloud provider and organizational context without breaking the AISVS mapping.
They do not just emit text. They drive CLAW for technical enumeration and return findings mapped to AISVS requirement IDs, with evidence you can sign. Not suggestions. Proof.
Every finding is evidence-backed, mapped to a requirement, and can be cryptographically signed into a tamper-evident record. Verification you can hand to an auditor, a board, or a regulator.
Run AISVS assessments with a method, not guesswork.
Attack AI systems and map every finding to the standard.
Verify the AI your developers and their assistants ship.
Turn AISVS into evidence and findings that hold up.
Deliver paid AISVS assessments at speed.
The toolkit behind the auditor credential.
Licensed to security teams and certified AISVS auditors. Private and proprietary. Drives CLAW and ships with the agent-skills and AI-DAST engine.
Enterprise and sovereign licences, on request.