🔍 Budget for the Checking: AI Does the Heavy Lifting, but Somebody Still Has to Look It Over

Or: Judges Weigh AI Prompts as Evidence, Employers Told to Save Chatbot Histories, Reviewers Busier Than Ever, an AI Coding Tool Leaks Passwords, and Clients Ask Where the AI Savings Went: This Week's eDiscovery AI News in Plain English

By Thomas Seest, Dunlap.Today · Tuesday, November 10, 2026

When the feed store got its first computer register, Mr. Hale didn’t throw out his pencil. At closing time he still ran the tape by hand, because the machine was fast but it didn’t know when somebody had keyed in a 50-pound sack as a 5-pound one.

That’s this week’s eDiscovery story. AI tools are doing more of the reading, sorting and drafting. But the checking still has to happen, it takes real time, and more and more often it ends up as part of the record. Courts are asking what people typed into AI. Clients are asking where the savings went.

Here’s the rundown: judges deciding when AI prompts become evidence, employers told to save chatbot histories, why checking AI’s work costs more than people budget, document reviewers busier than ever, an AI coding tool leaking passwords, new AI plug-ins, and clients asking why bills haven’t shrunk.

⏱️ What this means for you in 30 seconds

• Using AI on a case or with an expert? Keep the prompts. A judge may ask to see them.
• Something might turn into a lawsuit? Add AI chat histories to the list of things you save.
• Getting an AI-powered quote? Look for a line item for checking the AI’s work. If it’s missing, ask why.
• Building tools with AI? Make sure no passwords are sitting in files that get posted online.
• Paying legal bills? It’s fair to ask how AI changed the cost.

💬 What You Type Into AI Can Become Evidence

A law firm rounded up this year's court rulings on whether the questions people type into AI tools can be demanded by the other side. The short answer: it depends on who used the tool and why. One judge said an expert's AI prompts are part of how she reached her opinions, so the other side gets to see them.

A law firm reminded employers that what workers type into AI chatbots can become evidence, just like email. If a manager asked a chatbot for help writing up an employee before firing them, those chats could come out in a lawsuit.

Many court orders in lawsuits now say you can't put the other side's confidential documents into unapproved AI tools. A legal expert asked: what happens if someone does it anyway? Nobody knows yet, so lawyers are being urged to write the penalties into the order up front.

A University of Michigan student was disciplined for supposedly using AI on her papers. She says her disability makes her writing look machine-like. A federal judge threw out most of her claims but let her discrimination claim go forward.

It's like the notes in the margin of a seed catalog. Usually nobody cares. But if there's ever a dispute about what you ordered and why, those scribbles are the first thing folks want to read.

What it means for you: Treat AI prompts and chats about a dispute like any other case file: save them, think before you type, and agree on the rules with the other side early.

👉 Read more (Cozen O'Connor via JD Supra) · Read more (Amundsen Davis via JD Supra) · Read more (EDRM) · Read more (The Volokh Conspiracy)

✔️ Somebody Has to Check the Work

A consultant points out a cost hiding in AI investigations: somebody still has to check the AI's work. An AI summary can be 90% right and wrong about one key date, and you can't tell which part is wrong without going back to the documents. So budget for the checking.

A&O Shearman's eDiscovery team says AI isn't always the right tool. Keyword searches and simple filters still matter because everyone can see how they work. And AI 'search assistants' find the most relevant documents, not necessarily all of them.

An eDiscovery company co-founder says AI hasn't put document reviewers out of work. Each case needs fewer people, but cheaper review means many more cases are worth reviewing. And people are still needed to check the AI's results.

A UK law firm made entirely of software, which uses fixed rules rather than chatbot-style AI, helped a man get his visa after his first adviser picked the wrong category. It charges a flat £149 and hands people to a real solicitor when needed.

A new hired hand can bale twice as much hay, but you still walk the field to make sure he didn't leave a gap by the fence line. That walk takes time, and it belongs in the plan.

What it means for you: When AI speeds up a review or investigation, ask who checks the results, how much of it they check, and how much that costs.

👉 Read more (Ankura via JD Supra) · Read more (A&O Shearman via JD Supra) · Read more (EDRM) · EDRM on X · Read more (Legal Futures)

📋 Plans, Protocols and Paper Trails

An eDiscovery software company argues that buying more tools doesn't help if people still pass work between them by hand. Things like tracking who acknowledged a legal hold need a set, organized process.

An ESI protocol is the agreement in a lawsuit about how electronic records will be gathered and handed over. A software company shared a checklist and five reasons to use the same template every time.

Federal prosecutors are going after companies that lie about imports to dodge tariffs. HaystackID, an eDiscovery company, launched a service to check each customs filing against the paperwork behind it, with experts reviewing the AI's results.

Before the church fish fry, somebody writes down who brings the cornmeal, who runs the fryer and who counts the money. Nobody remembers the year it went smoothly, but everybody remembers the year it didn't.

What it means for you: Write down the discovery plan before the case gets busy, and make sure the handoffs between people and systems are tracked.

👉 Read more (eDiscovery Today) · Read more (eDiscovery Today) · Read more (ComplexDiscovery)

⚠️ Where AI Can Leak or Lead You Astray

A popular AI coding tool has been accidentally saving passwords into a settings file that sometimes gets posted publicly online. Researchers found about 1,100 such files with real secrets in them. If your team builds tools with AI, check your code storage.

A UK researcher warns that chatbots tend to agree with whoever is using them. In emotional family disputes, that can make people dig in harder. It can also help bad actors bury courts in convincing-looking paperwork.

Leaving the spare key under the mat is handy until somebody posts a picture of your porch online. And a friend who always agrees with you is pleasant company, but a poor advisor in a family fight.

What it means for you: If your team builds tools with AI, check for passwords in shared code. And don't let a chatbot's agreement stand in for good advice.

👉 Read more (eDiscovery Today) · Read more (Legal Futures)

🔌 New Plug-Ins for Legal AI

Litera, whose document-comparison tool most big US law firms use, is plugging it into Microsoft's Copilot for lawyers. The comparison itself is still done by fixed rules, so the AI summary sits on top of a result you can trust.

Harvey, a major legal AI company, now lets lawyers in Australia and nearby countries pull in research from publisher Wolters Kluwer without leaving Harvey, with links back to the original source.

A case-management software maker for injury lawyers added AI that builds a timeline of a client's medical treatment and drafts the demand letter to the insurer. Every fact links back to the record it came from, and a lawyer approves it.

A new attachment for the tractor is only as good as how well it hooks up and whether you can see what it's doing back there.

What it means for you: When an AI tool plugs into research, documents or case files, look for links back to the source so a person can check each answer.

👉 Read more (Artificial Lawyer) · Read more (Harvey) · Read more (Assembly Software)

💵 Money, Surveys and What Clients Are Asking

Blue J, a Canadian company that makes AI for tax professionals, raised $176 million. Its AI 'agents' can review tax returns and draft responses to IRS letters, showing the steps they took.

EvenUp, which sells AI to personal-injury lawyers, says it now has $150 million a year in signed contracts, double last year. One law firm signed a $16 million-a-year deal.

A Bloomberg Law survey found that one in four lawyers worries AI will take their job. Among younger lawyers it's more than one in three. Most also worry about AI 'going rogue.'

A survey of 800-plus law-firm leaders found nearly half have had clients ask why bills haven't gone down now that firms use AI. Many firms don't track profit closely enough per case to answer.

When the co-op got a faster grain dryer, farmers figured drying charges ought to come down a little. It's a fair question, and the co-op that can show its numbers keeps its customers.

What it means for you: Expect clients to ask how AI changed your costs. Firms and providers that track checking time and savings per matter will have the easiest answer.

👉 Read more (LawNext) · Artificial Lawyer · Read more (EvenUp) · Business Insider · Read more (Bloomberg Law) · Read more (BigHand) · EIN Presswire

✅ Checklist for this week

• ✅ Keep prompt histories for any AI an expert or litigation team uses on a matter
• ✅ Add AI chat histories, uploaded files and usage logs to legal hold notices
• ✅ Put specific remedies for AI misuse into protective orders
• ✅ Add a separate, priced verification step to every AI-assisted review or investigation budget
• ✅ Don’t rely on RAG search alone where completeness matters; keep search terms and filters in the mix
• ✅ Update your ESI protocol template to cover AI-assisted review and AI-generated data
• ✅ Scan code repositories for leaked passwords and API keys, especially AI tool settings files
• ✅ Be ready to show clients how AI changed hours and fees on their matters

Practical rule: Let the machine do the reading, but put the checking in the budget and on the calendar, and write down who did it.

— Dunlap.Today

📚 Plain-English AI Glossary

• Prompt: The question or instruction a person types into an AI tool.
• Expert methodology: The steps an expert witness used to reach an opinion, which the other side is usually allowed to examine.
• Litigation hold: A notice telling people to save records that may matter in a lawsuit.
• Protective order: A court order setting rules for how confidential case information can be used and shared.
• RAG (retrieval-augmented generation): AI that pulls a handful of the most relevant passages from your documents before answering. It’s helpful, but it can miss things.
• ESI protocol: The agreement in a case about how electronic records will be collected, searched and handed over.
• Deterministic: A system that follows fixed rules and gives the same answer every time, unlike chatbot-style AI.

❓ Frequently Asked Questions

Can the other side see what I typed into ChatGPT?

Sometimes. Courts are deciding case by case. An expert’s prompts have been treated as part of their methodology, while some people preparing their own cases have had their AI work protected. Ask your lawyer before using AI on anything tied to a dispute.

Does my company need to save employees’ AI chats?

If a lawsuit or investigation is reasonably expected, relevant AI chats, files and logs may need to be preserved just like email. Find out which AI tools people actually use, including personal accounts.

Doesn’t AI make document review cheaper?

Often faster, but someone still has to check the results. That checking takes real time, so a fair budget shows it as its own line.

Is AI putting document reviewers out of work?

One provider says no. Each case needs fewer reviewers, but more cases are now worth reviewing, and people are needed to check AI’s work.

Should my legal bills go down if my lawyers use AI?

It’s a reasonable question to ask. Nearly half of firms in one survey said clients are already asking it. A good firm should be able to explain how AI changed the work and the cost.

Other Resources

• Cozen O’Connor: Drawing the Discovery Line on AI prompts
• Amundsen Davis: Are AI Chats Discoverable?
• EDRM: Will Courts Enforce Protective Order Restrictions on AI Use?
• Ankura: Nobody Has Budgeted for the Checking
• A&O Shearman: AI in eDiscovery, Knowing When and How to Use It

Got an eDiscovery or legal AI question from the porch? Send it our way. We’ll keep translating the jargon into plain English.