AI
AI and LLM features
Prompt injection, output handling, personal data, keys and spend limits.
Scoped: runs under a signed scope letter
Who it is for
CTOs and heads of AI or product
Access we need
Test accounts and read-only code, under a scope letter. The public key check needs none.
What we check
- Provider keys in public JavaScript (reported by type and location only)
- User text reaching a model that can use tools (prompt injection)
- Model output rendered as HTML or executed
- Personal data sent in prompts, and prompt logging
- Per-user rate and spend limits, and fallback behaviour
- Mapping to the OWASP Top 10 for LLM Applications (2025)
Evidence you receive
Request and response pairs, injection fixtures and code locations.
How we rate severity
Critical for a provider key in a public bundle, or injected text that triggers an action on another user's data. High for no per-user spend cap.
What a fix looks like
Pull requests with tests: injection fixtures and spend-cap tests.
Included in
- Full Stack (Full Stack Audit)$19,900
- Also sold on its own: AI Feature Audit$5,900