AI controls in a managed IT environment: twelve topics.

Inventory of AI in use on managed devices and in the business tenant

Covers approved tools, AI features switched on inside existing software, and unofficial tools visible on managed equipment. Producing an inventory generally requires a discovery scan of the managed environment. Tools on unmanaged equipment are covered under topic 06.

AI features added to existing subscriptions

Software vendors have added AI features to existing products by update, among them assistants in office suites, CRM, accounting and help-desk tools. Whether such a feature is enabled is determined by the vendor’s default or by an administrator’s setting.

Vendor data terms and administrator settings

A vendor’s published terms state how it handles customer data. An administrator console exposes settings such as retention, training opt-outs and tenant boundaries. An IT provider can read the former and configure the latter; neither shows a vendor’s internal processing. The contract itself is a matter for counsel.

AI acceptable-use policy

An acceptable-use policy names approved and prohibited tools and sets data-handling rules. The policy is set by the business. Information about which tools are in use and which can be controlled comes from the systems the provider manages. Elements such policies commonly include are listed in FAQ Q.36.

Technical controls on AI tools

Common control categories are DNS and browser filtering, endpoint policy, tenant settings and application-approval workflows. Each applies only to the devices, networks and accounts it is configured on.

Personal devices and home computers

Controls applied to managed devices and networks do not reach personal phones or home computers. Measures that apply regardless of device include limits on which data staff can reach, the acceptable-use policy, and staff training.

Disclosure duties for customer-facing AI

Article 50(1) of the EU AI Act, applicable from 2 August 2026, requires AI systems that interact directly with people to be designed so that those people are informed they are interacting with AI, unless that is obvious from the context. A customer-facing chatbot may be built, hosted and operated by different parties: a web developer or agency, a software vendor, the business itself. Which party holds a duty in a given case is a legal question.

Regulation (EU) 2024/1689, Article 50(1) · applies 2 Aug 2026

Documentation at insurance renewal

AI exclusion endorsements have been filed with state regulators, with effective dates that differ by state, and they generally attach at renewal. Records that managed systems can export include an AI inventory, control settings and policy deployment records. What any policy covers is determined by its own terms, which this register does not read.

AI Insurance Filings Index, as of 28 Aug 2026

Visibility of unapproved AI use on the managed network

Detecting unapproved AI use depends on the logging and alerting in place. Some detection requires additional tooling, at additional cost.

AI in software vetting

Software evaluations can include questions on whether a vendor trains on customer data, which model provider it relies on, and whether its outputs carry provenance marks.

Agreement scope

Inventory work, policy support and additional controls may fall outside a standard IT support agreement. The agreement sets which are included and which are separately scoped.

Responsibilities of the business

Decisions on risk appetite, ownership of the acceptable-use policy and budget sit with the business. The provider implements technical controls within the agreed scope.