Dipo Thompson (@DipoThompson) header
Dipo Thompson (@DipoThompson) profile picture

forever learner. founder. legal tech

Des Moines, IA
Joined January 2020
First-level legal review is the biggest line in your discovery budget. Per-gigabyte and hourly pricing make it an invoice you can't forecast. Bring it in-house and a representative matter costs about half and runs 20x faster. discoveryai.io/resources/eco…
4
4
126
Privilege is not a name in the header. It is what the document says. Our privilege agent reads the content, not just who is on the email, and gives its reasoning. Your lawyers make the final call. discoveryai.io/resources/pri…
1
1
9
Who owns a lawyer's correction? Trove Report notes that two startups helping legal AI vendors train their own models say the edits people make inside an app are valuable training data. Harvey, a customer of both, says it never trains on customer data, and that deserves credit. In document review, every overturned call carries the client's facts and the reviewer's judgment, sometimes on privilege. That belongs to the client. Ask vendors in writing whether your edits train a model, and whose. It's one reason we run review inside the client's own Azure tenant. trovereport.com/p/harvey-won…
1
1
6
General counsel at Lloyds Banking Group and Compass Group told a panel at the Chambers Advantage Index UK launch that law firm invoices aren't falling far enough, even as firms invest heavily in AI. Clients are now asking firms to show the savings. That pressure lands first on routine, high-volume work like first-level document review. In-house teams have another option too. They can bring that first pass inside, keep the documents in their own environment, and send outside counsel the judgment calls that really need them. The AI helps, the lawyers decide. newlawjournal.co.uk/content/…
1
1
7
Pennsylvania's Supreme Court AI Advisory Committee has proposed a notice that would have every court filer certify they checked any GenAI-assisted content. Citations must exist, quotes must match the source, and stated facts must have a basis in the record. Comments are open until November 24. This is the right direction. The certification puts responsibility where it already sits, with the person who signs. It also rewards tools that show their work. When every AI suggestion comes with the source it relied on, checking it becomes part of the workflow instead of an afterthought. The AI helps, the lawyer decides. pacodeandbulletin.gov/secure…
1
6
A California appeals court let an AI-assisted expert report stand in Gillespie v. Superior Court, deferring to the trial judge's finding that ChatGPT was only a "thought partner." The court still called the evidence troubling. The report had a glaring factual error, and the expert couldn't produce her prompts because she deletes everything after use. Our take. Using AI to organize work is fine. Losing the record of how it was used is the real problem. When AI touches work a court will rely on, keep the trail and have a human verify it. Credibility alone shouldn't be the safeguard. edrm.net/2026/11/even-with-a…
6
An update on Fontainebleau v. Botach, the Florida AI chats case I wrote about last week. The magistrate judge has now ordered the defendant to finish producing his responsive, nonprivileged AI communications, log anything withheld, and consent to the AI providers disclosing deletion logs for a two-year period. The court made no finding of spoliation. The detail that stands out to me is that his ChatGPT and Grok account data went to an eDiscovery vendor for review. AI chat histories are becoming a review set like email, and every privilege call in them still needs a lawyer behind it. law.com/dailybusinessreview/…
1
1
11
A Connecticut federal court just added a gen AI clause to the protective order in Air Evac v. Aetna. Confidential discovery material can go only into secure, attorney-only eDiscovery platforms or enterprise tools that don't train on, retain or share the data. Anything else counts as disclosure to a third party. I think that's the right line, drawn around where the data goes, not whose logo is on the tool. The one part I'd change is naming vendors as examples. Courts should set the standard and let each party show it meets it. caselawwire.com/federal/air-…
1
1
7
Keyword searches find words, not meaning. So reviewers read the noise and the real documents slip past. Our agent reads first, judges each document against your request, and cites why. Your reviewer confirms or overrides. discoveryai.io/resources/wha…
1
1
10
The AI review pricing debate keeps fixating on the per-document headline. The number that matters is in the fine print. In the Winter 2026 ComplexDiscovery and EDRM survey, 39.6% of respondents didn't know how failed or exception documents are billed in AI review. Of those who did, 18.9% said exceptions go to manual review at standard rates. Per-doc, per-GB, credits and committed volume can all work. But if a legal team can't predict the bill before review starts, the unit is wrong. Our view is one number per matter, agreed up front, with exceptions spelled out. legaltechcircle.com/insights…
1
3
14
The Guardian reports UK public bodies are buried in AI-written complaints and information requests, many now ten times longer than before. The ICO says a person still has to answer them. Response teams need review that records why each item was released or withheld. theguardian.com/technology/2…
1
1
10
The New York Times reports that Crosby, an AI law firm, bought malpractice insurance for its AI agents, while major insurers are still working out how to cover AI risk. Insurance tends to follow accountability. Underwriters will likely want what courts already want, a record of why each call was made and who approved it. nytimes.com/2026/11/07/busin…
1
1
12
Courts keep drawing the same line on AI filings. In Farhat v. Cutera, counsel used a generative AI drafting tool for a first draft, caught citation errors in Westlaw, then filed the wrong PDF by mistake. The N.D. Cal. court skipped sanctions for that honest error, and still refused to consider the replacement authorities. My take. The tool is not the problem. The last check before the filing hits the docket is. AI can draft. A lawyer still has to own what goes out, the same way a lawyer should own every review call. edrm.net/2026/11/honest-erro…
1
1
12
Tiana Van Dyk at Epiq Canada puts the real shift in plain terms. Generative AI moved case insights to the start of review, not the end. Teams can spot a missing custodian or a scope gap before the budget is spent. That only helps if someone owns the process. Precision, recall, and elusion testing still matter. A lawyer still decides what leaves the door. canadianlawyermag.com/news/f…
1
1
10
Kelly Twigger's Case of the Week lands the right rule. AI prompts are ordinary ESI. They are discoverable when they go to a claim, a defense, or damages. They are protected only when they meet the same work product test as any other document. A Florida magistrate let a hotel dig into a defamation defendant's prompts because Florida's uncapped punitive damages turn on specific intent to harm. The final post shows what was said. The prompt history can show why. Treat the process like evidence from day one. Put AI accounts on the hold. Know what was asked and what was kept. minerva26.com/when-prompts-s…
1
1
13
Estonia's PM put a clean test on agents at Tallinn this week. Before an agent prepares a tax return or pays a bill, know what it is, what it is allowed to do, what it has done, and who is responsible. Litigation review needs the same four answers. An agent can make the first pass. Every call should carry a cited reason. The work should be able to run in the client's own environment when that is what the matter requires. The lawyer still owns what leaves the door. complexdiscovery.com/tallinn…
1
27
A&O Shearman just committed $1.1 billion over five years to AI, with each practice putting at least 3% of billable time into building specialist tools for its own area of law. That is a useful signal. The edge is not a generic chat box. It is domain expertise baked into systems lawyers can stand behind. Litigation review is one place that shows up clearly. First-pass coding across a matter should run with a cited reason on every call, inside the client's own environment when that is what the matter needs, and with a lawyer deciding what goes out. aoshearman.com/en/news/ao-sh…
1
1
1
44
EDRM's Rachi Messing makes a point the industry keeps missing. AI was supposed to end document review. Reviewers have never been busier. Faster, cheaper review does not shrink demand. It expands the set of matters worth reviewing properly. Fewer people per matter, more matters, and more work on validation, QC, and standing behind the process when the other side asks how it was done. That is the design we optimize for. Run in the client's own Azure tenant, put a cited rationale on every call, and leave the decision with the lawyer. edrm.net/2026/11/ai-was-supp…
1
1
2
289
Reveal's Onna can now collect Claude Enterprise chats, same as ChatGPT and Gemini. Above the Law flags the real issue. Plenty of employees still treat those prompts as private. They are not. Those threads are ordinary ESI, easy to search and produce, and often hard to explain. Privilege may apply sometimes, but banking on it later is a weak plan. My take. Collecting AI chats is overdue. Training people that every prompt is a record is more overdue. Treat the chat like email, and keep a lawyer deciding what gets kept and produced. abovethelaw.com/2026/11/ai-c…
2
3
102
Ontario's draft AI-evidence rules choose judicial control over prohibition. The quieter shift is reliance-based disclosure, which a Civil Rules Committee member says moves the e-discovery burden earlier in the case. Early, defensible review matters more. lawtimesnews.com/practice-ar…
1
30