By the Shadow AI Policy team
**A single ChatGPT session can breach information barriers that took years to build — and in consulting, that's an existential risk.** Consulting firms operate on trust. Clients share strategies, financials, and competitive plans because they believe those details stay contained. AI tools threaten that containment in ways most firms haven't fully addressed: the same consultant working two competing accounts can inadvertently cross-contaminate client context through a shared AI session, a personal account's conversation history, or a prompt that references one client while drafting a deliverable for another. This post covers the five AI governance challenges specific to consulting environments — information barriers, AI-assisted deliverables, partner oversight, AI strategy due diligence, and personal account risk — and what your policy needs to say about each one.Your AI policy isn't just an internal HR document — it's a client trust document. If a client asked to audit how your firm uses AI tools during their engagement, your policy is the first thing you'd hand them. Make sure it holds up to that scrutiny before they ask.
By the Shadow AI Policy team
Most AI policy guidance is written for companies with one set of confidential data — their own. Consulting firms have a fundamentally different structure: a single consultant may be simultaneously engaged with clients who are direct competitors, and the firm's value proposition depends on keeping those relationships completely siloed. That's a harder problem than most acceptable use policy templates are designed to solve.
The risk isn't hypothetical. A consultant drafting a market entry strategy for Client A might describe the competitive landscape in a prompt — naming competitors, pricing dynamics, and strategic weaknesses — without realizing that the AI platform stores that conversation, that their account is shared across devices, or that a colleague on a competing engagement has access to the same workspace. Standard enterprise AI tools aren't built around law-firm-style ethical walls. That's the firm's problem to solve, not the vendor's.
For a grounding on how unsanctioned AI tool use spreads inside organizations generally, see our overview of what shadow AI is and why it matters. The consulting context adds a layer that most of that general guidance doesn't address: your confidentiality obligations run to multiple external parties simultaneously, not just your own organization.
An information barrier (also called an ethical wall) is a structural separation between teams working on conflicting engagements. In law firms and investment banks, these are legally required in specific situations. In consulting, they're typically contractual — many client agreements include non-disclosure provisions that effectively require them, even if the word "barrier" never appears. Your AI policy needs to treat AI tools as a vector that can breach those barriers, because right now, most tools are not configured to respect them.
The core policy requirement here is account and workspace separation. Consultants assigned to engagements with competing clients must use separate, project-specific AI accounts or workspaces — not a single personal account that spans all their work. This is non-negotiable if you have clients in the same sector who would reasonably object to their strategic information being in the same AI session history as a direct competitor's. Some enterprise AI platforms (Microsoft Copilot with tenant isolation, for example) support workspace-level separation; others do not. Your policy should specify which tools are approved for which engagement types.
Policy language to consider:
The deliverable question has two distinct sub-problems: accuracy and disclosure. On accuracy, AI tools confidently produce analysis that looks authoritative and isn't. In consulting, that's a professional liability issue, not just a quality control issue. A slide deck with a fabricated market size figure or a misattributed competitive fact, delivered to a client CEO, reflects on the firm — regardless of how it was generated. Your policy needs to require human verification of any AI-generated factual claim before it goes into a client deliverable.
On disclosure, the question is whether clients have a right to know their deliverables were AI-assisted, and if so, how that's communicated. There's no universal legal requirement to disclose AI use in consulting deliverables today, but client contracts increasingly include provisions about it — and even where they don't, a client who discovers undisclosed AI use after the fact may reasonably feel misled. The safe default is transparency: build a standard disclosure approach into your engagement templates, and let partners decide when to go beyond the minimum, not whether to meet it.
Practical policy requirements for deliverables:
That last point is critical. Many AI tools train on or retain user-submitted content unless enterprise data agreements explicitly prevent it. Before your team pastes a client's three-year financial model into any AI tool, someone needs to have read that tool's data policy and confirmed it doesn't expose client data. Our guide on building an AI acceptable use policy covers how to structure those tool approval tiers.
In most consulting firms, AI tool adoption is happening bottom-up: junior consultants and analysts are finding tools that make them faster, using them on client work, and — in many cases — not telling anyone. This is the shadow AI problem, and it's acute in consulting because the stakes of an unreviewed tool being used on confidential client work are higher than in most industries.
The governance gap is usually at the partner level. Partners sign client contracts and are accountable for engagement quality, but they're often the last to know which AI tools their teams are actually using. Your policy needs to close that gap by making partner-level sign-off part of the AI tool approval process, not just an IT or Legal function.
Partners who sign client engagement letters are accountable for how work is produced on those engagements. AI tool selection is a delivery decision, not just a technology decision — treat it that way.
Specific governance requirements that work in practice:
A growing share of consulting engagements now involve advising clients on their own AI adoption — tool selection, governance frameworks, vendor assessment, implementation planning. This creates a conflict-of-interest risk that most firms haven't addressed in policy: a consultant who uses AI tool X personally may not be the most objective advisor on whether the client should adopt AI tool X, and in some cases the firm may have commercial relationships with AI vendors that should be disclosed.
Your AI policy should include guidance for consultants operating in an AI advisory capacity. This isn't just about objectivity — it's about the quality of the advice. A consultant who hasn't actually worked through their own firm's AI governance challenges is poorly positioned to advise a client on theirs. Make internal AI fluency a prerequisite for AI strategy engagements, not a nice-to-have.
For AI advisory engagements specifically:
Personal AI accounts are the single highest-risk vector in consulting firms. A consultant using their personal ChatGPT account to draft a proposal is not subject to enterprise data agreements, not covered by your firm's AI governance controls, and not generating logs you can ever review. Their conversation history may include client names, strategic details, financial figures, and competitive assessments — stored in an account that belongs to them personally, not the firm.
The risk compounds when consultants move between firms. Conversation history in a personal AI account travels with the person. A consultant who leaves your firm and joins a competitor takes that history with them. This is a genuine competitive intelligence exposure that has no clean analog in pre-AI policy frameworks, and most firm confidentiality agreements weren't written to address it explicitly.
The policy position here should be unambiguous: no client work on personal AI accounts, ever. This includes free-tier accounts on platforms that also have enterprise versions — the account tier matters because data handling terms differ materially between free and enterprise tiers. Make this a named prohibition in your policy, not a general principle employees are expected to interpret. To understand how broad the shadow AI problem typically is in organizations before a policy is in place, the policy generator at Shadow AI Policy can help you build firm-specific controls quickly.
Additional controls worth including in policy:
| AI Account Type | Data Retention Risk | Firm Visibility | Approved for Client Work? |
|---|---|---|---|
| Personal free-tier account (ChatGPT Free, Claude Free) | High — may be used for training; history follows the user | None | No |
| Personal paid account (ChatGPT Plus, Claude Pro) | Medium — training opt-out available but not default; still personal account | None | No |
| Firm enterprise account (ChatGPT Enterprise, Claude for Enterprise) | Low — zero data retention for training per enterprise terms | Admin-level logging available | Yes, with engagement-level controls |
| Firm enterprise account on conflicting engagement | Low for external exposure — risk is internal cross-contamination | Admin-level logging available | Only with workspace separation verified |
About Shadow AI Policy: We build AI acceptable use policy tools for HR and operations teams at 50–500 person companies. We publish guides on shadow AI, acceptable use policies, and AI governance, updated as regulations and AI tools change.
The most common mistake is treating AI policy as an IT or security issue rather than a client relationship issue. Consulting firms have confidentiality obligations that run to multiple external parties simultaneously, and an AI policy that only addresses internal data security — without addressing information barriers, deliverable disclosure, and engagement-level controls — misses the most consequential risks. Start with your client contracts, not your tech stack.
There's no universal legal requirement to disclose AI assistance in consulting deliverables in most jurisdictions today, but this is changing. More importantly, many client contracts now include provisions requiring disclosure, and even where they don't, clients who discover undisclosed AI use after the fact may treat it as a breach of trust or professional standards. The practical answer: review every new engagement contract for AI provisions, and build a default disclosure approach into your engagement templates now rather than handling it case-by-case.
Assign them to separate teams with no overlap, use separate firm-provisioned AI workspaces for each engagement, and document that separation in writing before the engagements begin. The responsible partners should sign off on the information barrier setup, including which AI tools are approved for each engagement. If complete team separation isn't operationally possible, escalate to firm leadership — this is a conflict management decision, not an AI policy decision.
Yes, but with specific guardrails. Don't upload client-provided confidential materials — strategic plans, financial data, competitive assessments — to AI tools during the advisory process. Disclose any commercial relationships the firm has with AI vendors being evaluated. And confirm that the consulting team has direct practical experience with AI governance, not just conceptual knowledge. Advising a client on AI governance while your own firm's practices are undocumented is a credibility problem, not just a policy gap.
Tailored to your industry and the AI tools your team uses. Free preview, $79 one-time or $149/mo with monthly updates.
Generate my policy kit →