Six AI Sanctions Decisions, One Discharge: What Actually Determines the Outcome

The six decisions, and the distance between them

Four circuit decisions and two district decisions between February and April 2026. Sanctions ranged from $15,000 per attorney with fees, double costs and a referral, through public reprimand and admonishment, to a $2,500 penalty, to case removal with denial of compensation, to nothing at all ^[1].

Laid side by side, the range is wider than most firms assume. These are not distinguishable by jurisdiction or by era; they are contemporaneous decisions from courts applying the same body of authority to the same category of conduct.

The heaviest is Whiting, where the Sixth Circuit imposed attorneys' fees, double costs, $15,000 in punitive sanctions on each attorney, and a disciplinary referral ^[1]. Multiple consequence classes in one order, which is the pattern our analysis of Johnson v. Dunn identified as the direction of travel.

United States v. Farris, also from the Sixth Circuit, is the one that introduces a consequence firms rarely model: alongside removal from the case and disciplinary referrals, the court denied Criminal Justice Act compensation ^[1]. Under the CJA, appointed counsel in federal criminal matters are compensated from public funds ^[4]. Denial means the lawyer did the work and was not paid for it, which is a sanction that never appears in a fine schedule.

At the other end, Gamez v. County of Fresno produced no sanction. The Eastern District of California discharged its order to show cause in April 2026 ^[1]. Same category of conduct, no penalty.

Fletcher v. Experian sits low on the monetary scale at $2,500 under Federal Rule of Appellate Procedure 46(c), which authorizes a court of appeals to discipline an attorney for conduct unbecoming the court or for failure to comply with applicable rules ^[3][5]. In re Nwaubani drew a public admonishment for conduct the court described as seriously interfering with the administration of justice, and Fivehouse v. U.S. Department of Defense drew a public reprimand ^[1].

Candor mitigates, and it is not enough on its own

The survey's central finding is a pair, and firms tend to hear only half of it. Attorney candor when confronted with errors operates as a mitigating factor. Candor alone does not prevent sanctions. Both halves matter, because a lawyer who expects an admission to end the matter has misread what the mitigation actually buys ^[1].

The first half is the part worth acting on immediately. Whatever else is true on the day a court asks about a citation, candor improves the position. It is the single variable most within the lawyer's control at that point, and it is free.

The second half is the corrective. Admitting the error does not make the sanction go away; it changes where in the range the outcome lands. A lawyer who is fully candid may still be reprimanded, still be referred, still pay. What candor appears to affect is severity, not existence.

That distinction matters for how a firm counsels a lawyer in the first hour. Advice framed as "tell the court and this will be fine" sets an expectation the record does not support, and a lawyer who is surprised by a sanction after being honest is a lawyer who was badly advised. The accurate framing is that candor is the cheapest available mitigation and it is not a defense.

It also aligns with what the disciplinary track rewards. Our analysis of the second proceeding at the state bar found the same pattern: the response after discovery carried more weight than the governance before it. Two different forums, applying different standards, converging on the same variable.

Evasiveness is the aggravator that costs the most

The survey is explicit that evasiveness significantly aggravates penalties ^[1]. That is the mirror of the candor finding and the more consequential half, because evasiveness is a choice made under pressure, usually by someone who has not decided in advance how they will respond.

The pattern recurs across the reported AI-sanctions cases well beyond this survey. The Ninth Circuit's decision in LNU v. Blanche turned substantially on the combination of fabricated authority and a lack of candor about its source. Del Biaggio v. Bansen sanctioned counsel for the misstatements and for a correction that arrived more than three months after notice.

The reason evasiveness is so expensive is that it converts one problem into two, and the second is worse. A fabricated citation is a competence and verification failure. Concealing or minimizing it once asked is a candor problem, and candor problems go to fitness rather than carefulness.

It is worth being precise about what counts as evasive, because most of it is not deliberate deception. Answering a narrower question than the one asked, describing the error without describing how it happened, attributing it to a process rather than a person, and delaying a correction while the firm decides on a position all read as evasive from the bench even when nobody intended to mislead.

There is a structural reason lawyers slide into it. The instinct that serves you well in litigation, which is to concede no more than the question strictly requires, is precisely wrong here. A show-cause order is not a discovery request from an adversary; it is the court asking a question about your conduct, and answering it narrowly reads as management rather than candor.

The practical implication is that the response should be decided before it is needed. A lawyer improvising under a show-cause order at 6pm is the person most likely to shade an answer, and the shading is what the court will weigh.

The discharge: what a court does when it imposes nothing

Gamez v. County of Fresno is the most useful decision in the set precisely because nothing happened. The Eastern District of California issued an order to show cause and then discharged it in April 2026 without imposing a sanction ^[1]. A discharge is not an acquittal, and it is not luck.

Courts issue show-cause orders to require an explanation, not to announce a penalty. The order is a question. A discharge means the answer was sufficient.

That reframes what the response is for. A lawyer receiving a show-cause order is not primarily defending against a sanction; they are answering a question about how this happened and what has been done about it. Those are answerable in a way that "please do not sanction me" is not.

What the survey's broader findings suggest a sufficient answer contains is consistent across the set: candor about what occurred, an account of the cause rather than a description of the symptom, and evidence of remediation. Johnson v. Dunn made the same point from the opposite direction, treating an internal review and an independent outside investigation as reassuring enough that suspension was unnecessary.

The caution is that a discharge in one case tells you nothing reliable about another. Facts differ, judges differ, and the survey itself notes that courts are signaling that lighter sanctions may be drawing to a close ^[1]. A lawyer planning on a discharge is planning on the least predictable outcome in the set.

The useful takeaway is narrower and more durable: the range genuinely includes zero, and where a case lands within it is substantially determined after the filing rather than by it.

No new rule is coming, and that is the finding

The survey concludes that existing rules sufficiently address Gen AI misconduct ^[1]. Courts are not waiting for an AI-specific sanctions rule, and firms should stop treating the absence of one as a gap. The authority already exists and is being used.

A recurring hope in firm discussions is that the profession is in an interim period before proper rules arrive, and that current enforcement is therefore improvised. The pattern across these decisions says otherwise.

Each of the six rests on ordinary authority. Fletcher applied Federal Rule of Appellate Procedure 46(c), which lets a court of appeals discipline an attorney for conduct unbecoming the court or failure to comply with the rules, and which long predates generative AI ^[5]. Farris used case removal and denial of compensation, both ordinary case-management powers. Public reprimand and admonishment are conventional appellate tools.

That matches what courts have said explicitly elsewhere. The Ninth Circuit reached its conclusion through Rule 11 and inherent authority, and the Johnson v. Dunn court proceeded under inherent authority precisely because it found Rule 11 inapplicable, without suggesting anything was missing from the rulebook.

The consequence for planning is that there is no rule change to wait for and no grace period to rely on. A firm's exposure is governed by rules that already existed, applied by courts that have now had two years of practice applying them.

The survey's other signal points the same way: courts are indicating that lighter sanctions may be drawing to a close ^[1]. Whatever leniency attached to novelty is being withdrawn as the conduct stops being novel.

Consequences that never appear in a fine schedule

Across these six, the monetary figure is frequently the smallest part. Denial of Criminal Justice Act compensation, removal from a case, public reprimand, admonishment, double costs, opposing party's fees, and disciplinary referral all appear, and most of them have no fixed number attached ^[1][4].

Firms that model AI-citation risk as a fine are modeling the least significant variable, which is a point this site has now made from three separate directions.

Farris is the clearest illustration. Denial of CJA compensation means appointed counsel performed the work and was not paid from the fund that exists to pay it ^[1][4]. That is not a penalty in the ordinary sense; it is the withdrawal of a payment, and it does not appear in any tally of AI sanctions organized by dollar amount.

Removal from a case carries a client consequence as well as a professional one, and the client did nothing wrong. Double costs and the opposing party's fees scale with how far the litigation ran rather than with the severity of the fabrication. A disciplinary referral, as our analysis of the disciplinary track sets out, starts a second proceeding with its own timeline and its own remedies against the license.

Public reprimand and admonishment are the ones most often dismissed as symbolic, and they are the most durable. A reprimand in a published decision is retrievable by name indefinitely, which makes it the consequence most likely to be found by a prospective client or a hiring committee years later.

Set against that, the monetary components are the ones a firm can absorb and forget. A $2,500 penalty under FRAP 46(c) is a line item; a public admonishment for conduct seriously interfering with the administration of justice is a description of the lawyer that stays attached to their name ^[1][5].

The asymmetry is worth stating plainly to anyone modelling this risk. The costs that can be quantified are the ones that end, and the costs that cannot be quantified are the ones that persist. A risk register that captures only the first has inverted the actual exposure.

Two years in, the leniency is being withdrawn

The survey's closing signal is that courts are indicating lighter sanctions may be drawing to a close ^[1]. That is a statement about time rather than about conduct. Whatever allowance attached to generative AI being new is being withdrawn now that it is not.

Early AI-citation cases carried an implicit discount. The technology was unfamiliar, the failure mode was not widely understood, and courts imposing modest fines were addressing something the profession had not yet absorbed.

That reasoning has an expiry date built into it, and these six decisions sit past it. By early 2026 the risk had been the subject of published appellate opinions, state bar guidance, court standing orders, and a great deal of professional commentary. A lawyer filing unverified authority in 2026 is not doing something the profession failed to warn them about.

The Johnson v. Dunn court made the same move in aggravation, noting that the attorneys had benefitted from repeated warnings, internal controls, and firm policies before the failure happened anyway. Knowledge of the risk is now assumed, and an argument that leans on novelty is arguing against a premise no court still accepts.

The practical consequence is that older sanctions figures are a poor guide to current exposure. A firm calibrating its risk on the modest fines reported in the first wave is using a baseline the courts have signaled they are moving away from, and the six decisions here already show the spread widening at the top.

It also means the mitigation available has narrowed to conduct. Novelty is no longer a mitigating fact, unfamiliarity is not a defense, and what remains within the lawyer's control is the response, which is where this survey locates the difference between outcomes in any event ^[1].

Decide the response before you need it

Every variable the survey identifies as decisive operates after the filing: candor, promptness, the quality of the explanation, and evidence of remediation ^[1]. All of them are degraded by improvisation, and all of them can be decided in advance.

The uncomfortable implication of a dataset like this is that the most consequential decisions are made by a lawyer under pressure who has never thought about the question before.

A firm that has decided in advance who is told, who investigates, who speaks to the court, and what the standing instruction is about candor has removed the improvisation that produces evasive answers. Our guide to the AI incident response plan covers the mechanics; this dataset is the argument for why the plan changes outcomes rather than merely documenting them.

The standing instruction is the part worth writing down explicitly, because it is counterintuitive under pressure. Full candor, immediately, including about cause, is the position that mitigates, and partial candor that answers a narrower question than the one asked is the position that aggravates.

1. Treat a show-cause order as a question, not an accusationGamez was discharged. Courts issue these to require an explanation; a sufficient answer ends the matter. Answering the question asked is a different exercise from arguing against a penalty ^[1].
2. Be candid immediately, and about the causeCandor is a mitigating factor and it is the cheapest one available. Describe how it happened, not only what happened; an account that omits the mechanism reads as evasive even when nothing is hidden ^[1].
3. Do not expect candor alone to prevent a sanctionThe survey is explicit that it does not. Advising a lawyer that honesty will resolve it sets an expectation the record does not support, and it affects severity rather than existence ^[1].
4. Correct on the day you knowDelay while the firm settles on a position is read as evasiveness. Del Biaggio weighed a correction that arrived three months after notice, and that timing was part of the sanction.
5. Bring remediation with you, ideally independentEvidence that the cause has been addressed is what converts an explanation into a sufficient answer. An outside review carries weight an internal one does not.
6. Assume no rule change is comingThe authority already exists and courts are signaling that lighter sanctions are ending. There is no interim period to wait out ^[1].