Adversarial validation for legal work

Every other legal AI helps you write the brief. LawNet's job is to break it before the judge does.

A private workspace built to attack your draft the way opposing counsel will, over whatever your firm already drafts with. Live today, the citation front pulls the actual case behind each citation and catches the fabrications that get lawyers sanctioned. The fronts that argue the other side and check the doctrine come online with the founding cohort.

In development, with a founding cohort of firms. The citation front is in production today.


The adversary the pipeline never had

Every legal AI is a straight line. Nothing in it ever attacks the work.

Retrieve, then draft or summarize, then cite-check. Every major product in this space shares that one shape, and nothing in that line ever turns around and tries to defeat what was just written. A drafting tool checking its own output is grading its own homework: the auditor cannot also be the bookkeeper.

LawNet is not a better step in that pipeline. It is the adversary the pipeline never had. A Stanford study measured the leading legal research tools producing hallucinated content in roughly seventeen to thirty-three percent of queries, even as their makers marketed them as hallucination-free. Verification has to come from outside the thing that wrote the draft.


The loop

Submit the draft. The swarm attacks it. You keep the record.

01

Submit

Bring the draft in from whatever tool wrote it. LawNet is model-agnostic: it validates output, it does not compete to produce it.

02

The swarm attacks

The citation front works the draft today, pulling and checking every citation. The opposing-counsel and doctrine fronts join the attack as they come online. Every finding traces to a real source the machine actually pulled.

03

Fix and re-attack

Accept or reject each proposed edit, then run it again. The draft is hardened until it survives the attack.

04

Keep the record

You get a validation report and a timestamped record that the work was checked: a record of what was examined, not a score and not a clearance.

Suggest-only, always. Every change is a proposed edit you accept or reject. LawNet never edits or files anything on your behalf. You remain the author of record.


The swarm

Three attack fronts, one job: find what opposing counsel would.

Citation checking is the floor. The front that argues the other side is what makes the brief stronger, not just safer.

The citation front

Live today

Pull the case, not a guess

Pulls the record behind every citation and confirms the case exists at that citation and is the case the filing names: the check that catches fabricated authorities, including ones squatting on a real citation's slot. It does not reason about whether a case is real: it looks it up.

Coming next Quote and holding verification: that the language quoted is really in the opinion, and that pin cites land where claimed. Whether the case truly stands for what the brief says is flagged for your review: a screen, not a verdict.

The opposing-counsel front

Coming online

Argue the other side

The strongest counterargument, the objection that lands, the thin spot in the record, and the case they will cite to distinguish yours. Simulated opposing counsel, before the real one files.

The doctrine front

Coming online

Check the standard

Whether the standard of review fits the court and posture, whether the elements are complete, and whether the burden is allocated correctly.

What a finding looks like. The report speaks one color per state, learned once and read at a glance.

Validation report Illustrative · not real matter data
Verified
Brown v. Board of Education, 347 U.S. 483 (1954) Citation and quoted language confirmed against the reporter.
Not found
Varghese v. China Southern Airlines, 925 F.3d 1339 (11th Cir. 2019) No matching case found at this citation. Consistent with a fabricated authority.
Flag
Reliance on the cited authority may overstate its holding. Surfaced as a candidate for your review. A screen, not a verdict.
Neutral
Six additional citations imported. Not yet checked.

The checks in full

What each check does. And what it deliberately doesn't.

Three fronts is how we attack. Here is exactly what each check does today, and what it deliberately doesn't. The disclosed limits are the point: silence never means verified.

Suggest-only Every check produces findings, never an applied edit. You stay the author of record; nothing is changed for you.

No silent caps Every check reports what it could not examine, an unuploaded record, a citation outside the public corpus, a citator step we do not run, as an explicit gap. The checks with hard ground truth come first.

Citation verifier Partially live

Live today For every authority the brief cites, it looks the case up in the public case-law record, never from memory, and confirms the case exists and that the party names, court, and year match what the brief says. It tells you plainly when a citation falls outside the corpus it can reach.

Coming next Verifying that quoted text and pin cites actually appear in the opinion, and that the case genuinely stands for the proposition it is cited for. Unlocked by full-opinion corpus access.

Does not It is not a citator: on its own it does not tell you whether a case is still good law (overruled, vacated, superseded). That needs a licensed citator (KeyCite, Shepard's, BCite) we do not integrate. Its reach is the public case-law record: a citation living only in a proprietary Westlaw or Lexis reporter comes back as an explicit coverage gap to check by hand, never passed as verified.

Hallucination catcher Coming online

Does Checks that every represented piece of text is actually in its claimed source: quotes appearing verbatim, pin cites landing on the right page, statutory text matching the current version, and "the court held X" not being a paraphrase of something the source materially does not say. Where the citation verifier asks whether the authority is real and rightly used, this asks whether every quoted or paraphrased text is real, including against your own record and opposing filings when you upload them.

Does not If the record or exhibits are not uploaded, it cannot check record cites, and says so per assertion ("record not provided"), never a pass. It verifies fidelity to the source, not whether the underlying legal argument is correct.

Adversarial counsel Coming online

Does Simulates the adversarial process the brief will actually face: opposing counsel's strongest counterargument and the distinguishing case they would cite, the objection a skeptical judge flags first, where an argument outruns the record, and the weakest link in the chain of reasoning. Every point is anchored to real retrieved authority or a specific span of the record or opposing filing, never a generic "a court might disagree."

Does not It runs degraded without matter context: with no opposing filings or record, it produces only generic counterarguments, and flags them as such. It surfaces threats to the draft; it does not yet propose affirmative winning theories the brief has not made. Like every check, it suggests; it never rewrites.

Doctrine and standard of review Coming online

Does Verifies you are arguing under the right legal standard for your court and procedural posture: the correct standard of review (de novo, abuse of discretion, clear error), complete claim or defense elements, correctly allocated burden of proof, and a standard drawn from controlling authority in your jurisdiction rather than one smuggled in from another circuit. Every check is grounded in a retrieved controlling case, never the model's memory.

Does not It does not tell you whether a case is still good law, the same boundary as the citation verifier: that needs a licensed citator we do not integrate. It checks the standard, not the merits of how you apply it.


The stakes

You cannot police every associate. You can catch the citation before a judge does.

One hallucinated citation in one filing is enough. The sanctions, the malpractice exposure, and the reputational hit land on the lawyer who signed it, not on the tool that wrote it.

You cannot guarantee that no one at your firm is quietly using AI, and you cannot read every draft. What you can do is put an adversarial check between the draft and the filing, so a fabricated citation is caught inside your walls instead of in open court.

It started with Mata v. Avianca (S.D.N.Y. 2023). It did not stay small, and it did not stay with the consumer chatbots: lawyers using the professional legal AI tools have been sanctioned too.


The offense

Harden yours. Then break theirs.

Today the citation front checks your own draft. The swarm is built to point the other way, too.

When the adversarial fronts come online, point it at what opposing counsel filed: their weak citations, their overstated holdings, the counterargument they did not answer, and what is genuinely strong on their side, stated honestly, so you know where not to spend your time.

Twenty days before arbitration

A forty-page brief with fifty citations lands. Checking it by hand is two days of an associate's time. Or you run it through the swarm, and spend those two days on the argument instead.


Calibrated honesty

We under-claim on purpose.

Incumbents marketed hallucination-free and hallucinate anyway. We tell you exactly what was checked, and what was not. Under-claiming is the product.

Verified

A citation is verified only when the machine pulled the actual authority and it held. Not because it looks right, and not because a model is confident. Green is earned.

Flagged

A possible mischaracterization or a weak spot is surfaced as a candidate for your judgment. A screen that points you at what to examine, not a verdict.

Never

We will never tell you a document is safe to file. LawNet reports what it checked. Whether the brief is ready is your call, and it stays yours.


SciNet

From the makers of SciNet, the adversarial engine behind a living graph of scientific claims, where AI agents publish, review, and reproduce results in the open. The same engine, now being pointed privately at your practice.

Visit SciNet

Early access

We are building this with a founding cohort of firms.

LawNet is early. We are working with a small group of firms to ground it in real practice before opening it more widely. If liability from AI-assisted work sits on your desk, we would like to talk.

  • A working session on how your teams use AI today, and where the exposure is.
  • The citation front today, and the adversarial fronts as they come online, shaped by your matters.
  • Confidentiality by design: a workspace isolated to your firm, never used to train a model, never pooled across firms. Exact terms set with your security team.

Invite-only, for now

We are onboarding a founding cohort of firms first. If someone sent you this link, reply to their message and we will get you set up.