Professional services firms have a specific AI governance problem that general enterprise guidance does not address well.
A software company that leaks proprietary code into ChatGPT faces a product risk. A law firm that does the same with client matter documents faces a privilege waiver, a potential regulatory complaint, and a client relationship that may not survive the conversation. The stakes are categorically different.
At the same time, AI tools are now deeply embedded in how professional services work. Associates use ChatGPT for research and first-draft memos. Consultants use Claude for deck preparation and data synthesis. Accountants use Copilot for financial analysis. The tools are faster than the governance frameworks. That gap is what this checklist is designed to close.
This guide is written for managing partners, heads of IT, and practice group leaders at law firms, management consulting firms, accounting practices, and financial advisory firms — particularly those operating under GDPR, working with enterprise clients who send security questionnaires, or preparing for an internal compliance review.
The specific risks for professional services
Before the checklist, it is worth naming why professional services firms face a harder version of this problem than other businesses.
Client data is in everything. In professional services, the work product is inseparable from the client matter. A consulting deliverable, a legal memorandum, a financial model — all of these contain client information by definition. When employees use AI tools to produce or accelerate that work, client data is inevitably in the context.
Privilege and confidentiality are not the same as GDPR compliance. A law firm might correctly believe that attorney-client privilege protects the substance of client communications. It does not protect against a GDPR violation if a staff member pastes personal data from a client file into an external AI model. The privilege and the compliance obligation are independent.
Your clients are now asking. As noted above, enterprise clients — particularly those in financial services, pharmaceuticals, and technology — are adding AI sections to their vendor questionnaires. Professional services firms, as vendors to those enterprises, are receiving those questionnaires. The question “how do you handle our data in your AI tools?” requires a specific, documented answer.
Regulators are paying attention. Bar associations in multiple jurisdictions have issued guidance on AI use in legal practice. GDPR supervisory authorities have begun scrutinizing AI-mediated data transfers. Firms that have not documented their AI controls are taking on compliance risk that is no longer theoretical.
The AI governance checklist for professional services
1. Inventory every AI tool in active use
Before you can govern AI, you need to know what AI is in use. This is harder than it sounds. AI is often adopted bottom-up — an associate tries a tool, finds it useful, and it spreads. By the time IT knows about it, it’s a workflow dependency.
What to do: Survey your teams quarterly. Ask specifically about browser-based AI tools (ChatGPT, Claude, Perplexity), AI-embedded applications (Copilot, Grammarly, Notion AI), and any AI tools used for client-facing work. The list will be longer than you expect.
Control checkpoint: Can you produce a current approved AI tools list? If not, you cannot answer a client security questionnaire.
2. Classify your data before it reaches the AI tool
Professional services firms typically work with multiple data categories that have different handling requirements: personal data (GDPR-regulated), privileged matter content (confidentiality obligation), commercially sensitive client information (contractual obligation), and internal firm data (lower risk).
What to do: Define which data categories may and may not enter which AI tools. A practical policy: public information and draft internal content → approved AI tools with logging; client matter content and personal data → restricted tools only, with explicit inspection controls; privileged communications → not to be submitted to any externally-hosted AI model.
Control checkpoint: Does your AI usage policy define permitted data categories per tool? Is it written, dated, and acknowledged by fee earners?
3. Enforce the policy at the technical layer, not just the training layer
The most common gap in professional services AI governance is the gap between the policy and the practice. “We told everyone not to put client data in ChatGPT” is not a control. It is a hope.
What to do: Deploy an AI gateway — a control layer that sits between your users and the AI tools they access, inspecting prompt content and enforcing data classification rules automatically. When a user attempts to submit content that matches a defined sensitive pattern (client name, matter number, PII), the gateway can redact it, block it, or route it for approval before it reaches the model.
Control checkpoint: Is policy enforcement automatic, or does it require employee memory? If an employee forgets the policy under deadline pressure, what happens?
4. Establish an approved tool list with an explicit review process
An approved AI tools list without a review process becomes outdated within months. AI tools change rapidly — capabilities, data handling terms, and data residency arrangements are updated frequently.
What to do: Establish a quarterly review cycle for the approved tool list. For each tool, document: the current data processing agreement (DPA) or equivalent, the data residency arrangement, whether personal data processing requires a GDPR Article 28 processor agreement, and any special restrictions for client matter use.
Control checkpoint: When was your approved tool list last reviewed? Do you have signed DPAs with every AI provider your firm uses?
5. Create an audit trail for AI-assisted work product
When a client or regulator asks “was AI used in preparing this document, and if so, what data was provided to the model?” — can you answer? In some jurisdictions, lawyers are already required to disclose AI use in certain filings. The question of what data the AI model received is a natural follow-on.
What to do: Ensure that your AI governance layer produces an audit trail that logs, per interaction: the user, the tool/model, the timestamp, the policy decision applied, and whether any sensitive content was detected or redacted. This log should be tamper-evident and retained for the period required by the applicable regulatory or contractual obligation.
Control checkpoint: For any AI-assisted work product you delivered in the last 90 days, can you reconstruct what data was submitted to the model?
6. Build a human approval gate for high-risk AI actions
Not all AI use is equal risk. A lawyer using AI to summarize publicly available case law is different from a lawyer using an AI agent to review and extract terms from a confidential client contract. The second case calls for human review before the output is acted on.
What to do: Define which AI actions require explicit human approval before proceeding: actions involving identified client personal data, actions producing client-facing deliverables, and actions by autonomous AI agents accessing internal matter management systems. This is especially important as firms begin using AI agents for document review and contract analysis.
Control checkpoint: Are there AI-assisted workflows in your firm where the AI output reaches the client or a regulatory submission without human review? If so, is that an intentional design choice or an oversight?
7. Address GDPR Article 28 processor obligations for every AI tool
Under GDPR, every tool that processes personal data on your behalf must be covered by a Data Processing Agreement (DPA) that meets Article 28 requirements. This applies to AI tools that handle EU personal data — which, in a professional services context, is nearly all of them.
What to do: For every AI tool in your approved list that may receive personal data: confirm whether a GDPR-compliant DPA exists, review the data retention and deletion terms, confirm data residency (where is data processed and stored), and verify that sub-processor disclosures are adequate.
Control checkpoint: Does your GDPR register include AI tools as processors? Have those processor relationships been reviewed in the last 12 months?
8. Document your AI governance for client questionnaires and audits
The checklist above generates the documentation that answers client questionnaires and auditor requests. The work is wasted if it is not documented in a retrievable form.
What to do: Maintain a single AI governance document or register that includes: current approved tool list, data classification policy with AI addendum, DPA status for each approved tool, log retention policy, and policy review dates. When a client questionnaire arrives, this document is your answer source.
Control checkpoint: How long would it take you to answer the AI section of a client security questionnaire today? If the answer is “more than an afternoon,” the documentation is not in place.
A note on GDPR and the “legitimate interest” trap
Some firms attempt to justify AI use of client personal data under GDPR Article 6(1)(f) “legitimate interest.” In a professional services context, this is usually insufficient. The typical legal basis for processing client personal data is Article 6(1)(b) (contract performance) or Article 6(1)(c) (legal obligation). Submitting that data to a third-party AI model provider typically falls outside the scope of those bases without explicit disclosure to the data subject.
Before deciding that existing GDPR grounds cover your AI use, have a qualified privacy lawyer review the specific data flows involved. The question is not whether you can justify AI use in general; it is whether you can justify submitting specific categories of personal data to a specific AI provider under the basis you have documented.
Getting from checklist to implementation
The checklist identifies eight controls. In practice, the hardest one to implement is number three — technical policy enforcement. The others can be addressed with documentation, process, and DPA reviews. Number three requires a technical layer.
For professional services firms that are not large enough to have a dedicated security engineering team, the practical path is a deployed AI gateway that covers browser-based AI access (where most AI use happens) and optionally desktop and mobile. The gateway enforces the policy you have written, produces the audit trail you need, and handles the DPA with one vendor rather than requiring a separate DPA with every AI tool your staff uses.
The goal is to reach a state where, when the next client questionnaire arrives, you hand over a document rather than a guess.
Qadar AI Shield is designed for professional services firms that need AI governance without a dedicated security team: deployed in a day, cross-provider policy enforcement, and a built-in audit trail that satisfies GDPR, client questionnaire, and regulatory review requirements. Learn more.