Nakul Pandya
83% of lawyers now use generative AI, but only about a third of firms have formally sanctioned it. That gap has a name: Shadow AI.
At many firms right now, a managing partner is "still evaluating AI tools" while their associates down the hall are pasting depositions into a free ChatGPT/Gemini or any other general purpose tool.
If you're a partner or managing partner at a boutique, regional, or midsize litigation firm, this post is about the real cause and effect of your team using general-purpose AI tools for litigation prep, deposition summaries, medical record chronologies, and the other document heavy work that lives inside your case files. The gap between what your firm has approved and what your attorneys are actually doing is wider than you think, and it carries risks your firm never signed up for.
What Shadow AI Actually Looks Like at a Midsize Firm
Shadow AI refers to employees using AI tools their employer hasn't approved, vetted, or even necessarily knows about. Running queries through general-purpose tools on personal accounts instead of firm-sanctioned platforms with proper data controls.
At a litigation firm, that looks like an associate uploading a 300-page deposition transcript to a consumer chatbot at 9 p.m. because the summary is due tomorrow. No one approved it. No one logged it. And your client's records just left the building.
There are two major issues that shouldn’t be neglected at any cost:
1. The data sensitivity problem
Client medical records, deposition testimony, and claim files are among the most sensitive documents in any industry. General-purpose tools offer no default data protection plans. Their stronger privacy protections are usually gated behind paid enterprise tiers, while a large number of attorneys are using free or consumer accounts, exactly where data handling terms are weakest.
Purpose-built legal platforms address this with SOC 2 Type II and HIPAA compliance and a zero data retention workflow meaning client data is never used to train the vendor's models or any third-party models. Your data stays your data.
2. Efficiency, accuracy and the context problem
General-purpose tools don't solve for efficiency, accuracy or case context. Associates end up in a loop: upload the document, refine the prompt, get a half-usable answer, edit it manually, re-upload the next day into a fresh chat that remembers nothing about the case. That's not productivity improvement
Technically speaking, these tools parse long documents in chunks, which means they routinely miss critical information, the kind of fact that can make or break a case.
A tool that generically summarizes a deposition isn't aware of the injuries at issue, the defense theory, or the client you're defending.
How a purpose-built platform compares against general-purpose AI tools. newcase.ai is designed and engineered around litigation work specifically. Here's what that difference looks like for a single feature: deposition summaries.
newcase.ai (context-aware) | General-purpose AI tools | |
|---|---|---|
Output quality | Clean, structured summaries that surface all key facts, page-line summaries, abstracts, and key admissions | Large blocks of generic text dumped on the user, not suited for litigation work |
Information search | Natural-language, context-aware search across the entire deposition. Ask anything, get the exact relevant testimony with relevance ranking | Keyword-level matching with no ranking |
Case context | A context-aware system that understands each case and produces summaries that are built based on the facts of the case | Every summary is done in the same way without taking case context into account |
Verification | Every fact is linked directly to its page-and-line citation in the original transcript, so verification is instant | Every output must be manually checked against the uploaded transcript |
Note-taking | Take notes and tags directly on the deposition transcript | No way to annotate the underlying document |
Sharing | Share summaries, page-line citations, and notes with colleagues in one click, or export to Word or PDF | Manual copy, paste, and reformat |
Deposition summaries are just one of the features attorneys have found the most value in. If you want to see the difference on one of your own transcripts, you can start for free or request a demo. A 300-page deposition takes about 25 seconds. We'd be happy to walk your firm through it.
Skepticism is just the delay
Just like any other technology, the privilege of using cutting-edge tools has always been restricted to a few early adopters until it becomes the norm.
That's exactly what's playing out with AI in the legal industry. While boutique and regional firms remain skeptical, the largest firms are already miles ahead: enterprise-wide generative AI adoption at large law firms is now approaching 100%, and 63% of midsize firms have formally adopted at least one gen AI tool.
And to be honest, the skepticism is absolutely fair. It's always difficult to believe in something that seems impossible at first. Someone who has never swum doesn't know the joy of swimming.
Here's a question worth asking: Why did your firm first adopt case management software? Weren't you skeptical then, too?
Every technology wave in legal looked optional right up until it wasn't:
Era | Inflection point | What changed |
|---|---|---|
1973–75 | Computerized legal research | Before Lexis and Westlaw, research meant physical law libraries |
Late 1980s–90s | The PC era | Billing software and document management entered the back office |
2006–2012 | E-discovery + cloud | Electronically stored information became formally discoverable; practice management moved to the cloud |
2023 | Generative AI | The first time technology could do substantive legal work, drafting, summarizing, analyzing rather than just organizing it |
Today | Litigation intelligence | Context-aware platforms like newcase turn entire case files, records, depositions, exhibits into a single searchable intelligence layer |
The firms didn't adopt any of these waves early. They adopted under pressure, new discovery rules, client demands, competitive necessity. The only question with AI is whether your firm moves before or after the pressure arrives.
Why Banning AI Makes Things Worse
The numbers tell the story: 83% of lawyers individually use generative AI, while only about a third of firms have formally adopted legal-specific tools.
The forbidden fruit effect.
It means the more firmly something is off-limits, the more tempting it becomes, and the more likely people are to reach for it in secret rather than give it up. Applied to shadow AI: the firm's "don't use unsanctioned tools" policy doesn't stop attorneys from quietly using general-purpose tools, whether or not those tools actually help them. The ban creates the shadow.
Prohibition doesn't stop the use. It just makes it invisible, leaving the firm with no control over data privacy, no visibility into how client materials are being handled, and no way to measure the actual ROI of the AI work already happening inside its walls.
Questions Firms Should Ask Before Picking a Solution
The way out isn't a stricter ban, it's giving your attorneys a sanctioned tool that's specifically built for tasks they’re doing. Before choosing, every firm should be able to answer these questions:
Does the vendor meet real security standards? Look for SOC 2 Type II and HIPAA compliance, and critically a zero data retention policy, meaning your and your client data is never used to train the vendor's models or anyone else's.
Does our firm have a clear set of guidelines on AI use? A one-page policy naming approved tools, usage guidelines and a person to ask beats a 40-page memo nobody reads.
Are we choosing a tool that not only does the work, but does it verifiably? In litigation, an answer without a citation is a liability. Every extracted fact should link back to its page-and-line source so verification takes seconds, not hours.
Does the tool keep experts in the loop? The right system augments attorney judgment rather than replacing it. AI does the review and ranking; your attorneys stay in control of conclusions, strategy, and decisions. For high-stakes matters, human-in-the-loop QA should be available.
Is it built for litigation, or built for everything? A general-purpose tool knows a little about all documents. A litigation intelligence platform understands depositions, medical chronologies, and expert testimony as connected parts of one case.
If a tool can't answer all five, it’s probably not a fit for litigation and legal work.
FAQs
Can lawyers use ChatGPT/Claude for client work?
Consumer tiers of general-purpose tools may retain and train on the data you input, which can put client confidentiality and privilege at risk. Several state bar ethics opinions now require lawyers to vet an AI tool's data handling before inputting client information. The safer path is a legal-specific platform with SOC 2 Type II and HIPAA compliance and zero data retention.
What is shadow AI in a law firm? Shadow AI is the use of unapproved AI tools by attorneys and staff, typically free or personal accounts of general-purpose chatbots outside the firm's knowledge or data controls. Surveys show 83% of lawyers now use generative AI individually, while only about a third of firms have formally sanctioned legal-specific tools, meaning most legal AI use today is shadow use.
How many law firms use AI in 2026? Individual adoption has reached 83% of lawyers according to Bloomberg Law's 2026 State of Practice survey. At the firm level, 63% of midsize firms have formally adopted at least one generative AI tool, and enterprise-wide adoption at the largest firms is approaching 100%.
What should a law firm AI policy include? At minimum: a short list of approved tools that meet security standards, a bright line on what data can never enter consumer tools (client identities, medical records, deposition testimony), basic training for every attorney, and a named person responsible for AI questions. The goal is governance, not prohibition, bans just push the use underground.
How is a litigation intelligence platform different from ChatGPT? General-purpose tools process documents in chunks, lose case context between sessions, and provide no citations. A litigation intelligence platform like newcase.ai reads every line, connects facts across records, depositions, and exhibits into one searchable layer, and links every extracted fact to its page-and-line source with accuracy validated against 100,000+ reviewed pages.
Will AI replace the associates doing this work today?
The firms getting real value aren't trying to. The model that works is Human + AI, not AI instead of humans: the platform accelerates document review and surfaces critical facts, while attorneys make every decision about analysis and strategy. What changes is where associate hours go, from re-reading transcripts to taking on more cases, thus benefiting the firm.
