Shadow IT is the old name for employees using software the firm never approved. Shadow AI is its newer, faster-moving cousin. An associate drafts with a personal chatbot account. A paralegal installs a browser extension that summarizes web pages. A partner turns on an AI meeting recorder that joins every call. Nobody is being reckless; they are trying to work faster. But each one may be sending client information to a vendor the firm has never reviewed.
If your firm has no AI policy, you almost certainly have shadow AI. If it has a policy that bans everything, you probably still do, only less visibly. The sensible response is to find out what is happening, assess the risk and offer better options.
Why people go around the rules
Understanding motives makes the fix easier.
- The approved tools do not exist yet, or are slower.
- People have used AI at home and see obvious value.
- Competitors and colleagues talk about it constantly.
- Nobody told them what is allowed.
- Deadlines create pressure to take shortcuts.
A policy that ignores these pressures is hard to enforce. One that offers a practical, safe alternative tends to work better.
How to find out what is in use
Ask first
The simplest approach is a short, anonymous survey. Ask which AI tools people use, for what tasks, and whether they have entered client information. Promise that the purpose is to understand needs, not to punish. Honest answers are the most valuable data you will get.
Check technical signals
Your IT provider can often review:
- Web traffic and DNS logs for visits to well-known AI sites.
- Browser extensions installed on managed devices.
- Installed applications on laptops.
- Cloud app discovery reports from security or identity tools, which show which third-party apps users have signed in to with firm accounts.
- OAuth permissions granted to third-party apps connected to Microsoft 365. Review which apps can read mail or files.
- Meeting platforms, where recording or transcription bots may have been added.
Technical review can't see everything, particularly personal devices, but combined with the survey it gives a decent picture.
Assess the risk of each tool
For each tool discovered, ask:
- Was client or confidential information entered?
- Is the account personal or business? What do the terms say about retention and training on inputs?
- Where is the data stored and who are the subprocessors?
- Is there an alternative approved tool that does the same job?
- Does any client agreement prohibit this use?
ABA Formal Opinion 512, issued in July 2024, notes duties of confidentiality, competence and supervision in the generative AI context. Those are the lens through which to rank the risk. Anything touching privileged material in a consumer account belongs at the top of the list.
Respond proportionately
High risk: stop and contain
If client data was entered into an unapproved consumer tool, treat it as a potential confidentiality incident. Document what was entered, check the provider's deletion options, and have a partner or ethics counsel decide whether client notice is warranted.
Moderate risk: bring into the fold
If a tool is useful and the vendor's terms are acceptable on a business plan, run it through your vendor review and approve it formally. Moving a tool from shadow to sanctioned is often better than banning it.
Low risk: guide
Some uses involve no client data, such as brainstorming a CLE topic. Say so explicitly in the policy.
Build a better sanctioned path
People stop using unofficial tools when the official ones are good enough.
- Provide at least one approved general-purpose assistant under business terms.
- Offer a short list of approved legal-specific tools for research and drafting.
- Publish a simple request process, with a fast answer, for new tools.
- Provide short training on safe use.
- Check in quarterly on what staff need that is missing.
Control the technical exits
Where appropriate, use technical controls: block categories of unapproved AI sites on managed devices, restrict browser extension installation, limit third-party app consent in Microsoft 365 and set data loss prevention rules to catch obvious sensitive patterns. Technology cannot replace judgment but it can reduce accidents.
Keep the culture safe for honesty
If the first person to admit mistakes is disciplined harshly, nobody else will. Make clear that reporting is expected and appreciated. Partners should model it by disclosing their own uses.
Review regularly
New AI features arrive inside software you already own. Revisit the inventory twice a year and after major product updates.
How Counsel Cyber can help
Counsel Cyber can run a shadow AI discovery for your firm, review the tools it finds and help you set up approved alternatives. If you would like to see what is actually in use at your firm, we are glad to help.