Trustwelt AI Security · Policies

One list of policies. The first that decides, decides.

A policy says, for the people it applies to, on the services and channels it names, what happens to each class of data. The first policy in the list that decides a class wins; what no policy decides follows your organisation's defaults. Legal may give confidential text to one approved assistant and nobody else; personal data typed into a translation service may wait for your answer; a new rule may only observe for a week before it bites.

A module of Trustwelt AI Security, included with it. Runs on: Managed computers with Trustwelt Desktop and its browser extension: the Chromium family and Firefox on Linux, Firefox on Windows; the AI agent gateway on managed Linux computers; other programs through the network filter, sites only.

Tenant Admin › AI services › Policies: the version in force, an ordered list of policies with who they apply to, their services and the action per class of data, and the organisation defaults.

What it does

Policies, in plain words.

  • Scope

    People, services, channels

    Everyone, or named people and access groups; every service or named ones; paste, drop, upload, send, or all.

  • Actions

    Five, per class

    Let in; let in and record; mask, then ask; hold for review until an administrator answers; keep out.

  • Observe

    Record what it would do

    An observe policy lets the text in and records the decision it would have made, so you see the effect before you enforce it.

  • Guardrails

    What no policy loosens

    Restricted data is always kept out; a secret is never let in as it is, even under observe.

  • Lifecycle

    Write, publish, pilot, roll back

    Changes stay unpublished until you publish them; a version can go to a pilot group first; every version is kept and any can be published again. No second approval is needed.

  • Start

    Templates and the simulator

    Start from Protect personal data, Protect secrets, Approved AI only or a Detect-only pilot; ask the simulator what would happen to chosen classes for a person on a service, and which policy decides.

How it works

From the console to the computer.

  1. Write

    Tenant Admin › AI services › Policies: add a policy or a template, order the list, set the defaults.

  2. Simulate

    A person, a service, a channel, the classes: the decision and the policy that made it, with no text involved.

  3. Publish

    To everyone or to a pilot group. Each computer gets its own statement at its next check.

In Tenant Admin

What your administrators see.

Tenant Admin › AI services › Policies: the version in force, an ordered list of policies with who they apply to, their services and the action per class of data, and the organisation defaults.
Policies: ordered, versioned, published to everyone or a pilot group.

What is recorded, and what never is

Your organisation sees this. Nothing else.

  • VersionsEvery published version with who published it and when; saved changes with before and after in the audit log.
  • DecisionsEvery input check names the policy version and the policies that decided it.

Never:

  • A policy that lets restricted data through or a secret in as it is: the guardrails hold whatever is written.
  • Text kept by the simulator: it works from the classes you choose.

Specification

Where it runs and how it is sold.

Needs
Managed computers with the current Trustwelt Desktop; an older Desktop applies the strictest action everywhere, and Coverage says so.
Limits
Up to 100 policies; every version kept.
Sold as
Part of Trustwelt AI Security.

Questions

Asked before buying.

Does a stricter policy lower in the list ever win?
No. The first policy in the list that decides a class decides it; you order the list. The guardrails apply on top, whatever the order.
Can I try a policy without stopping anyone?
Yes. Set it to observe, or publish the version to a pilot group first. Observe records what it would have done; the pilot applies it to the group alone.

More in Trustwelt AI Security

The other modules, and what works with this one.

Start with Trustwelt AI Security.

Policies comes with Trustwelt AI Security, sold on its own per person per month. An organisation account is free.