By ALIS LABS
Audit Copilot
Internal controls and audit with AI, designed to be auditable.
Context
Internal controls and audit share one problem: critical processes, manual work and scattered evidence. The cost is collecting, validating, explaining and recording.
Challenge
- Evidence in many formats: PDFs, spreadsheets, emails, screenshots.
- End-to-end traceability: source, analysis, conclusion.
- AI risk: hallucination, unsupported conclusions, prompt injection.
Solution
A platform with separate domains for controls, audit, evidence and workflow. Agents help design tests, organize evidence and draft conclusions, always with human review in the flow.
Architecture
- 01Domains: controls, audit, evidence and workflow
- 02API with security middleware and rate limiting
- 03Versioned database migrations
- 04Prompts versioned outside the code
Governance & security
- Guardrails declared in configuration, not scattered in code
- Automated red teaming with garak across the three model providers used (Claude, Gemini and GPT)
- Audit trail: who asked, what came out, from which sources
Current state
- In evaluation
- Operating figures will be published once real data exists