Keeping the Attorney in the Loop

How Litmas AI was Built to Support Your Judgment, not Automate It

Legal AI Tools Should Center Attorney Judgment

A purpose-built litigation AI tool designed with attorneys in mind should center attorney judgment. Consider a pre-AI workflow, in which a partner assigned an associate a task such as drafting a research memo analyzing a client’s tort claim or a motion to dismiss for failure to state a claim. When the partner reviewed that associate’s work, the partner may have assessed the associate’s judgment and stress-tested the memo. The partner might have asked why the associate cited a minor case instead of a major case, why the associate chose to focus the argument on count and not another. They might ask whether the underlying contract on which the motion to dismiss is based is authentic and not subject to challenges. These questions don’t go away just because an AI created the first draft (or an associate drafted it with AI). In fact, these questions become even more important when AI drafts the memo. An attorney must be able to query an AI platform in the same way as they would an associate or paralegal, and to receive a reasoned, sound, truthful response, not a sycophantic assessment of what the AI thinks the attorney wants to hear.

The full measure of a legal AI tool isn’t just what it automates, though that is an important feature. It’s also whether it allows the attorney to see and influence every important decision made along the way. Quality AI tools do not aim to completely replace an attorney, and they understand that there are decisions that must be deferred to the professional who has spent years in school and practicing.

Why the Attorney Has to Stay in the Loop

These professional expectations aren't new and should be familiar. It doesn’t matter whether it’s framed as the attorney’s duty of competence, duty of candor, or duty to supervise. It doesn’t matter if it’s Rule 11’s requirement that the legal and factual contentions of a filing are warranted, or if it’s an individual judge’s standing order on using AI. The answer is the same: the buck stops with the filing attorney. Not with the AI, not with the paralegal, not with the law student intern.

How this duty has manifested, though, has evolved with the introduction of new technology. Pre-AI, every law student has rightfully been terrified of citing law that’s no longer good in a legal filing and being called out about it by opposing counsel (or worse, the judge). Updating has been drilled into every young attorney. That’s the reason when a red flag shows up on a case cited in a filing during a final cite check, young lawyers go into spins. But, as demonstrated by the thousands of cases in which hallucinated, incorrect, or misrepresented information has found its way into legal filings, AI exacerbates and magnifies both the mistakes and the issues. The ABA’s Formal Opinion 512 reiterates the concept: the use of generative AI tools fall squarely within attorney’s already-existing ethical and professional responsibilities.

The solution is obvious and as simple as it’s aways been. Attorneys have to supervise AI work just as they supervised paralegal and intern work. California’s recently-passed law on attorney duties over AI makes the case plainly: attorneys can’t delegate the practice of law to a generative AI system (we’re glad California agrees with us). Again, this isn’t new. Attorneys have never been able to delegate the practice of law to a non-lawyer.

If this is just a different manifestation of the same duty, then, it begs the question: why do AI tools seem to make attorneys forget about that judgment? Why do so many attorneys seem to lose the judgment loop when they use AI in a way that they wouldn’t when reading over an associate’s work?  Why do they lose sight of the duties laid out in Formal Opinion 512?

Why AI Tools Cause Attorneys to Leave Judgment Behind

The same partner who would spend an hour grilling an associate on case selection will often accept an AI's first draft after a five-minute skim. Nothing about the underlying duty changed. What changed is the psychology of the interaction, and it breaks down in a few predictable ways.

First, we tend to see automation bias, where people extend more trust to a machine's output than to a colleague's. A memo with five cleanly formatted citations reads as authoritative, whether those cases say what the memo claims.

Second, there is a confirmation bias, where people look for evidence to support their belief rather than contradict it. In many generic AI tools, including Claude and ChatGPT, this bias is intentionally exacerbated, since these tools are built to satisfy the prompt, not challenge it. An attorney who drafts a prompt has already stated a theory of the case, and the generic AI tool obliges. A sharp associate might push back and argue that the strict liability claim is stronger than the negligence one and draft accordingly. A tool optimized for compliance won’t do that, because it’s built to please. An attorney who follows the ideas a generic AI might miss the stronger argument they would have caught had they drafted the memo themselves.

Then there's the anchoring bias, compounded by the fluency issue. With the anchoring bias, once a generic AI tool spits out a draft, it becomes the baseline for editing rather than a starting point for interrogation. That effect only gets stronger under deadline pressure. Layer on top of that the fluency problem, where generic AI-generated text reads as competent because it's fluent. But, as we know, fluency is not the same thing as soundness. An attorney reviewing an associate's memo can ask follow-up questions and watch the associate reason in real time. A static AI output offers no such surface to probe accurately, unless the tool was built to provide one. A well-worded output does not make it an accurate one. AI tools are abundant, and it’s easy for attorneys who are constantly under deadlines to look for fast and easy ways to get through work product. No one can blame them for that. But the tools, the deadlines, and the amount of work don’t change attorneys’ professional responsibilities. Generic AI tools not only weren’t built to help attorneys adhere to these responsibilities, they weren’t even built with them in mind.

Litmas AI Puts You Back in the Driver's Seat

Litmas AI is built around a simple premise: offer attorneys a way to utilize AI technology to help draft work product while still adhering to their ethical and professional responsibilities. Litmas AI (unlike Harvey, Legora, Claude, or ChatGPT) was purpose-built with the litigation case management process front, centered, and as the target goal. Every citation is source-linked and traceable, so verification isn't an extra step the attorney has to take outside the platform. Litmas AI’s Agentic Associate was built not to assume it knows best when a decision should be made by an attorney. It will point out gaps in an argument or suggest focusing on different, stronger argument. But even when that happens, it’s alerting the attorney and allowing them to decide. And, just like a good associate or the partner themself, Litmas AI has the context around the case, the law, and how you and your firm operate.

An attorney can query the Agentic Associate the same way they'd query an associate and get a reasoned, specific answer back. The tool is designed around a series of control points, moments where the attorney makes the call, not the AI.

Here are just a few ways Litmas AI Agentic Associate places the attorney solidly in the loop:

- Any research done by the Agentic Associate comes with a citation attached, so you can check the work at its source.

- Under Review Draft, the attorney can see the motion with linked case law and statutory references. That link will provide you with information about the case, including the full opinion itself.

- Any citation that Litmas AI pulls is verified and updated against our proprietary database of case law, statutes, and regulations before it enters your document, so you can spend time on judgment rather than tracking down potentially hallucinated cases or worrying about citing cases that are no longer good law. Additionally, Agentic Associate runs a good law analysis on each case.

- The Verification workflow in the system allows you to review your motion and click on any factual pincite. Each fact is hyperlinked to the place in the record. So, you have full visibility to confirm for yourself. That link will take you to the exact page cited.

- Every section of a document, at any stage, can be edited, approved, or regenerated before it goes anywhere. Nothing moves from draft to filing without an explicit decision from you.

This attorney in the loop and the verification workflow are the real distinction between a purpose-built litigation AI tool like Litmas AI and generic AI tools like ChatGPT and Claude. Generic AI tools weren't built around a duty to supervise and an understanding of attorney judgment. Litmas AI was. Co-founded by an experienced litigator, it was built for attorneys from the very first line of code. The attorneys reading this already understand that their judgment can't be delegated. The smart move isn't avoiding AI. It's choosing the system built by litigators who understood your ethical and professional responsibilities before they built anything at all.

Book a Demo Today.

FAQ

1. Does Litmas AI file or send without my review? No. Every draft stops at your desk before it moves anywhere else. You download and file from there.

2. Doesn’t Claude for Legal work the same way? No. Litmas AI was designed and built by litigators with a deep understanding of the litigation case management process. Other legal tools have been strapped on to its existing AI, not initially and intentionally built for legal.

3. How is this different from tools that say most of the work is done before I look at it? Litmas AI is built around your review, not around minimizing it.

4. Does this change my duty to supervise the work? No. It gives you the visibility to do that supervision faster, at the fact and citation level instead of the whole-document level.

5. Can I change which facts or arguments show up in the draft? Yes, at any point, and the draft updates around your choice.

Remember: Litmas AI doesn't ask you to trust it. It gives you what you need to decide for yourself, fast enough that the review doesn't erase the time you saved