What is Shadow AI and why is it a critical risk for regulated firms?

Why Shadow AI is the risk your policies can’t see

Use Shadow AI here to mean AI use outside the firm's approved controls, including tools that are available but unmanaged. Inventory the tools, data and output uses, then assess which tasks need an approved route or a pause.

A visibility gap leaves questions to investigate: where data went, which prompts and model were used, what was returned and how outputs were checked before influencing decisions. Assess the available evidence and applicable requirements; do not infer a compliance breach or total loss of records from an incomplete inventory.

So what counts as Shadow AI?

Check for unapproved uses and for permitted tools whose data, configuration or review arrangements are unmanaged. The examples below are discovery prompts, not evidence that each occurs in the firm:

• Staff pasting client or customer information into a public AI chatbot
• Teams using AI note-takers or meeting transcription tools without a data protection review
• ‘Copilot’ style tools enabled by default inside productivity suites with unclear tenant controls
• Developers using AI coding tools connected to repos without policy, logging or approvals
• Third-party vendors using AI on your data without contractual visibility

AI Risk & Efficiency Audit

Why a ‘no AI’ stance can increase risk

A statement that the firm does not use AI is not a usage check. Compare the policy with discovery records and staff accounts before concluding whether use has stopped or continues outside approval.

Define permitted tasks, restricted data and an approval route staff can follow. Check whether the approved tools meet those tasks and whether staff understand the restrictions; do not assume availability alone changes behaviour.

Risk areas regulators care about

Assess the five areas below against the observed use. An unmanaged tool does not by itself establish a regulatory finding; have the relevant reviewers decide which requirements and risks apply:

• Data protection: assess the permitted processing, extent of sharing, any required DPIA and international processing arrangements
• Confidentiality: check whether client or sensitive internal information was disclosed and which restrictions apply
• Model risk: examine outputs for invented facts or unsuitable advice and check review before use
• Auditability: establish which decisions, prompts, data references, approvals and controls can be reconstructed, and record gaps
• Operational resilience: assess supplier outages, changes in policy or use, and tools outside the maintained inventory

How to get Shadow AI under control (without killing productivity)

1. Inventory reality: discover what is being used (browser extensions, SaaS, meeting tools, code assistants)
2. Classify data: define what can never be put into external models and what requires approval
3. Approved stack: provide a sanctioned toolset with clear guidance and logging
4. Controls: DLP, access controls, retention, monitoring and incident response playbooks
5. Training: practical, role-specific training that explains what “good” looks like

Compare policy with observed use, record the material gaps and assign owners. Prioritise the affected workflows by information, impact and unresolved controls, then check whether the selected changes address those gaps.

AI Managed AI Governance & Support

AI Implementation

Need help assessing an AI workflow and its unresolved risks?

For shadow AI discovery, identify the actual tools, accounts and material involved, then prioritise review using the consequences of each use.

• Audit AI risk and opportunity
• Build a usable AI governance framework
• Implement AI with supplier and data controls
• See AI support for financial advisers