Every firm with a restrictive policy and no good sanctioned tool has shadow AI. The only variable is whether it knows.
Start a conversation with the AI Adoption Concierge, already scoped to approved tools & shadow AI. Pick a starting point, or describe your situation directly.
Shadow AI is the use of unapproved tools for firm work, and it is the predictable consequence of a gap between what people need and what they have been given. The tools are free, they are capable, and they are one browser tab away on a personal device the firm does not manage. A policy that prohibits without providing an adequate alternative does not eliminate the usage; it removes the firm's ability to see it, govern it, or know what client material has left. Firms that treat this as a discipline problem tend to make it worse. Firms that treat it as a product gap — what were people trying to do, and what would they have used if we had offered it — tend to close it.
Almost never defiance. Almost always friction, a gap, or not knowing there was a rule.
Extra steps, another login, documents that must be moved. Friction paid dozens of times a day loses to a tab.
The register covers research but the need was summarising. People solve the problem they have.
The policy exists and was not read, which is a communication failure rather than a compliance one.
Work done on equipment the firm does not manage and cannot observe.
The rule is known and set aside at midnight, which is exactly when the risk is highest.
The firm cannot tell compliance from non-compliance, so both look identical from the centre.
How firms close the gap.
The confidentiality risk is the serious one; the loss of visibility is what makes it unmanageable.
People reaching for an unapproved tool are telling you what the approved set does not do. Firms that respond by enforcing harder get quieter shadow use. Firms that respond by closing the gap get compliance almost for free.
Ask, in a way that carries no consequence, and pair it with whatever technical visibility the firm properly has. An anonymous survey stating plainly that the purpose is to fill gaps rather than to discipline anyone gets remarkably candid answers — people generally know they are working around something and would rather not. Network-level and identity-provider signals show which services are reached from firm systems, though not what went into them, and monitoring has its own limits and obligations that the firm should take advice on.
Establish the facts quickly — what material, which tool, when, and what the terms of that service permit regarding retention and training. Then take advice on the firm's obligations, which may include client notification depending on the material, the jurisdiction and any contractual commitments. Handle it as a confidentiality incident rather than a disciplinary one at first: the immediate priority is understanding exposure, and a response that leads with blame reliably ensures the next occurrence is not reported.
Rarely enforceable and frequently counterproductive. Firms with genuine mobile working cannot practically prevent someone opening a browser on their own phone, and a rule that cannot be enforced erodes the credibility of the rules that can be. The more effective controls are making the approved tools available on the devices people actually use, being unambiguous about what may never be entered anywhere, and ensuring the sanctioned option is good enough that reaching past it is not tempting.
Current enough that it does not itself create shadow use. A register listing tools the firm evaluated a year ago, with nothing added since, tells people the process has stalled and that asking is pointless. Firms that keep it useful review it on a short cycle — quarterly at first — and publish a route for requesting an addition with a commitment to answer within a defined period. The route matters as much as the list: an unanswered request becomes an unapproved tool.
Describe your approved set and what you can see. The Institute will help you close the gap.