Product

The path to agentic AI

People decide every answer. Only confirmed answers are scored.

AI proposes, a named reviewer decides.

Agentic AI, here, means AI that does the groundwork, under your review. This page describes the path. It is not a list of features you can use today.

The control principle

Software reads the evidence. People make the call.

These four steps are how AI-assisted suggestions are designed to be reviewed. The version now in testing follows them on test documents only. It is not used on client documents.

  1. AI reads the documents

  2. Proposes an answer, citing the page or clause

  3. A named reviewer accepts, edits or rejects each one

  4. Only confirmed answers are scored. The AI never scores.

The stages

From what is live today to the direction of travel

  1. Live today

    Browser assessment software, available today. You run it in your browser. Hosted live demo on sample data available upon request. The hosted version is not live for clients.

  2. Built, in testing

    1. Documents are loaded for review.
    2. The software reads the uploaded documents and suggests an answer to each question, quoting the exact passage and page it relies on. Where the documents don’t support an answer, it says so rather than guessing.
    3. A named reviewer confirms, changes or rejects each suggested answer. Only confirmed answers are scored.
  3. Next

    Across the whole assessment, on an EU-hosted AI model, with citations to the client’s uploaded documents. A named reviewer decides.

  4. Planned direction

    These are planned directions, not a schedule.

    • Improvement plan drafting and tracking
    • Board report sections for management review
    • Alerts when evidence goes out of date
    • Collecting documents from systems the client chooses to connect
    • Periodic re-assessment

Destination

Vision, not current capability

Agentic AI for risk and compliance: a program that stays assessed and documented, with people approving every decision.

Commitments

How review and data are designed to work

These are commitments for the hosted and AI path, not a description of what is available today. The browser version available today keeps answers on your device.

  • The AI never scores. Every suggestion is shown with its source and accepted or rejected by a named reviewer.
  • Every decision will be logged: who accepted, edited or rejected each suggestion.
  • Assessment data will be hosted in the EU, and client documents will be processed in the EU.
  • We will only use AI providers that contractually exclude training on client data.
  • Independent security review before any client documents are processed by AI.

What the assessment covers today, and what is planned

This is about the scope of the assessment, not about the AI path. Labels show what is available today and what is planned.

Today

Assessment keyed to the DOJ’s Evaluation of Corporate Compliance Programs and the FCPA Resource Guide, plus an anti-corruption add-on mapped to ISO 37001.

Planned

Planned future mappings to ISO 37301, the UK Ministry of Justice guidance on adequate procedures under the Bribery Act 2010, and the OECD 2021 Anti-Bribery Recommendation.

Compliance domains for AML and sanctions.

Mappings show where assessment questions relate to each framework. They are not a statement of compliance and are not endorsed by any authority. Mapping does not imply ISO certification.

See it in a demo

The live demo on sample data is open only by request.