AI Guard
See, govern and audit the artificial intelligence your organisation is already using
AI Guard gives an enterprise one control point for artificial intelligence use. It finds the models, assistants, plugins and agent features in use across your estate — including the ones adopted without approval — applies policy at the point of use, blocks the paths that leak information, and records what was sent, to which model and under whose authority. The result is adoption your risk function can sign off on: people keep working with the tools that help them, inside boundaries that hold when nobody is watching.
What it does
Discovery of shadow AI
Identify model providers, assistants, browser extensions and embedded AI features in use across the estate, mapped to the teams using them.
Policy at the point of use
Allow, warn or block specific models, features and data classes before a prompt leaves the device, rather than after it reaches a provider.
Data loss prevention for prompts
Detect and redact or withhold regulated, contractual and intellectual property content in prompts, attachments and agent tool calls.
Agent and automation controls
Scope what autonomous workflows may read, write and call, with approvals required for actions that touch production systems or customer data.
Vendor and model assurance
Track where model providers process and retain data, their training use, and your contractual position, so approvals rest on evidence rather than a sales deck.
Immutable audit trail
Prompt, response, model, user, purpose and policy decision recorded together, exportable for regulators, customers and internal review.
Risk and cost visibility
Usage, exposure and spend by team, model and use case, so governance effort and budget follow what is actually happening.
Aligned to existing frameworks
Controls mapped to ISO/IEC 42001, NIST AI Risk Management Framework, and the EU AI Act obligations your organisation already reports against.
What you get
- A complete inventory of AI use, including unapproved adoption
- Sensitive information stopped before it reaches an external model
- Board and regulator-ready evidence of who used what and why
- Faster, documented approval of new AI tools and use cases