How to Audit Your Firm for Shadow AI Usage in 48 Hours | Insights

You cannot govern what you cannot see. Before your firm can address the compliance crisis created by unsanctioned AI usage, you need to quantify it — who is using what, on which client data, and how often. This is the 48-hour audit protocol for IT Directors and Compliance Officers at regulated professional services firms.

For the full analysis of the scope of Shadow AI compliance exposure in Scottish firms, read our parent article. This piece gives you the operational methodology to measure the problem within your own perimeter.

Phase 1 (Hour 0–12): DNS Logs and Network Traffic

Every connection from a firm device to a public AI service leaves a trace in your DNS resolver and firewall logs. Build a watchlist of AI service domains and query against the past 30 days of outbound traffic:

OpenAI: chat.openai.com, api.openai.com, cdn.oaistatic.com
Google Gemini: gemini.google.com, gemini.googleapis.com
Anthropic Claude: claude.ai, api.anthropic.com
Microsoft Copilot (consumer): copilot.microsoft.com
Perplexity: perplexity.ai
Summarisers: quillbot.com, writesonic.com, jasper.ai

Filter by internal source IP. Sustained POST traffic to api.openai.com indicates active data submission — someone is feeding documents into an inference endpoint, not casually browsing. Cross-reference each source IP against your DHCP lease table or endpoint management platform (Intune, Jamf) to identify the individual. You need names, departments, and roles.

Phase 2 (Hour 12–24): Browser Extensions and Email Forensics

DNS logs capture browser-based AI usage. They do not capture AI-powered browser extensions, which operate silently within the browser process and transmit page content to external APIs without the user typing a single prompt.

Export a full extension inventory from every managed browser. Chrome Enterprise, Intune, and Jamf all provide centralised extension reports. Flag any extension requesting Read and change all your data on all websites — this permission grants the extension access to every page the user views, including your practice management system and client correspondence.

Simultaneously, search your firm’s email infrastructure for registration confirmations from AI platforms. Query corporate email accounts for messages from noreply@openai.com, noreply@anthropic.com, and noreply@google.com containing “welcome,” “subscription,” or “receipt.” Review corporate card statements and expense claims for charges to “OpenAI,” “Anthropic,” or “AI subscription.” Any match is a confirmed indicator of Shadow AI adoption the employee considered work-related enough to expense.

Phase 3 (Hour 24–36): Structured Staff Interviews

Select 8–12 individuals across departments: client-facing staff, support teams, and anyone handling high volumes of client documentation. Conduct brief, non-punitive, fact-finding interviews. Five questions:

“Have you used any AI tool for work tasks in the past three months?”
“What types of tasks?” (Drafting, summarisation, research — each carries a different data exposure profile.)
“Did you paste or upload any client documents or correspondence into the tool?”
“Which platform, and was it a personal or firm-provided account?”
“Are you aware of any firm policy on AI tool usage?”

Document every response. These records form part of the compliance evidence file.

Phase 4 (Hour 36–48): The Shadow AI Risk Register

Consolidate all four data sources — DNS logs, extension inventories, email/expense indicators, and interview records — into a single Shadow AI Risk Register. For each confirmed instance, record the individual, the tool, the account type, the data category exposed, the frequency, and the specific UK GDPR Articles engaged (Article 5(1)(f) for security failure; Article 28 for uncontracted processor; Article 44 for international transfer).

This register is the document your Compliance Officer needs to brief the Managing Partner. It is also the document the ICO would expect to see during an investigation — evidence that the firm identified the risk, quantified it, and took corrective action.

After the Audit: Detection Without Remediation Is Worse Than Ignorance

An audit that identifies Shadow AI and is followed by no structural change is evidence of known risk and deliberate inaction — the precise aggravating factor that escalates ICO penalties. The firm must issue an immediate directive prohibiting unapproved AI tools, and simultaneously deploy a compliant private alternative that delivers the same productivity gains with zero data leakage.

The administrative controls buy time. The infrastructure solves the problem.

Talk to our team to see how a zero-egress, bare-metal sovereign AI deployment eliminates Shadow AI risk at the architectural level.

Private AI Ltd (SC881079)