Professional Responsibility
What the AI era demands: competence, supervision, and candor.
Opening Argument
One argument, drawn from the week to date.
This week the Idaho Supreme Court called one litigant's use of AI-fabricated citations "pervasive and egregious" in a unanimous opinion, then affirmed her $1.1 million defamation judgment without adding a single dollar in sanctions. That gap between language and consequence was not unusual. Of ten AI-hallucination rulings entered by U.S. courts in the seven days ending September 23, three ended in a warning, three opened a show-cause inquiry still awaiting a response, three closed with no AI-specific penalty at all, and exactly one, a pro se filer in Indiana, produced a dollar figure: $23,222 in costs.
Read together, the bench's message is that AI fabrication is now taken seriously in principle and handled case by case in practice, with the paper trail rarely closing the loop on what happens next. None of this week's sharpest judicial language attached to a member of the bar; the two federal show-cause orders naming lawyers, in Pennsylvania and the District of Columbia, are still open.
Into that gap has stepped an unlikely source of an actual standard: malpractice insurers, who told firms this week, in specific terms, to put a written AI policy in place and verify everything, because a policy is the one document an insurer, and eventually a court, will actually ask to see.
The Implication
Do not wait for a court to define your standard of care. Write the AI policy your malpractice carrier is already describing, and treat the two pending show-cause orders as a preview of what an unwritten policy costs once it reaches your own filing.
The Findings
What moved in practice this week, and what each one tells you.
Idaho's Supreme Court called a litigant's AI-fabricated citations "pervasive and egregious," then declined to add any sanction to her $1.1 million judgment.
↗ Reason, The Volokh Conspiracy
In a unanimous September 22 opinion in Posey v. Bushnell, Chief Justice Richard Bevan wrote that only six of twenty-two case-law quotations in Summer Bushnell's appellate brief were accurate, and that the record raised "serious doubt" about her denial that she used AI. The court affirmed the verdict against her and warned future litigants of possible sanctions, but imposed none here.
Why it matters This is the highest court yet to put that warning in writing, but against a pro se litigant with no bar card at risk. The next test is whether a state supreme court applies the same language, and an actual sanction, to a licensed lawyer.
An Indiana appeals court affirmed $23,222 in costs against a pro se litigant who cited six fabricated cases in a property dispute.
↗ AI Hallucination Cases Database
In Englert v. Gateway Crossing Property Owners Association, decided September 18, the Indiana Court of Appeals let stand a costs order tied to briefing that relied on six fabricated citations and four more it misrepresented. It is the only ruling among this week's ten AI-hallucination cases to carry an actual dollar figure.
Why it matters A costs order is a consequence a self-represented litigant actually feels, unlike a warning. Watch whether appellate courts default to costs, not warnings, once a fabrication has already been flagged once below.
A federal judge in Philadelphia ordered a lawyer to show cause over eight fabricated or misrepresented citations in a civil filing.
↗ AI Hallucination Cases Database
In Whitfield v. Prime Agency, entered September 18 in the Eastern District of Pennsylvania, the court identified two fabricated cases, three invented quotations, and three misrepresented holdings, and ordered counsel to respond before it rules on sanctions. No decision has been published yet.
Why it matters Unlike Englert, this lawyer holds a bar card and a Rule 11 obligation, not just a losing brief. How the response is drafted, and whether it draws a disciplinary referral, is the harder half of this story.
A District of Columbia federal court opened a second lawyer-filed show-cause inquiry the same week, over false quotations in an employment case.
↗ AI Hallucination Cases Database
In Garner v. Transit Employees Federal Credit Union, entered September 22 in the U.S. District Court for the District of Columbia, the court flagged three false quotations attributed to case law and ordered a response. As with Whitfield, it is pending.
Why it matters This is the third lawyer-filed show-cause order in seven days, alongside Whitfield and an Illinois probate matter, In re Estate of Ruth M. Gerdes. Track these dockets for the eventual sanction, not the order itself, before concluding courts treat counsel harder than pro se filers.
Malpractice insurers are telling firms, in specific terms, what an AI policy needs to contain, filling a gap no bar has yet made mandatory.
Published September 21, the piece quotes Berkley Select's Al Roberts, ALPS's Chris Newbold, L Squared Insurance's Lee Norcross, and CiteCheck AI's Tom Martin on the same five asks: a written AI policy, independent verification of every output, controlled data environments, staff training, and a clear line that the lawyer, not the tool, owns the judgment.
Why it matters No bar has made all five mandatory yet. An insurer asking for them at renewal works the same way a rule would, and it will arrive at the next renewal cycle, not on a regulator's timeline.
The Practice
This week supplied ten data points on the same question: what actually happens after a court catches AI-fabricated citations. The clearest single answer came from the Idaho Supreme Court, the highest court so far to rule squarely on it, and that answer complicates any easy lesson.
Sorted by outcome, the week's ten rulings split into four groups: three warnings, three open show-cause orders, three closures with no AI-specific penalty, and one costs order with a dollar figure attached. No two courts reached the same ending by the same reasoning.
10
AI-hallucination rulings entered by U.S. courts in the seven days ending September 23, by outcome.SOURCE · AI HALLUCINATION CASES DATABASE, DAMIEN CHARLOTIN
On the Methodology
One researcher maintains this database from public filings and reader submissions; it lists 2,077 cases to date but does not claim to be exhaustive, and it likely undercounts matters resolved without a written order. The outcome labels are the database's own classification, not a uniform standard applied the same way across every court.
Every ruling above came from a court, not a bar, and the two tracks move at different speeds. A judge can enter a show-cause order within days of catching a fabricated citation, as Pennsylvania and D.C. did this week. A bar can turn a fabrication complaint into a license consequence, but on a much longer clock: the State Bar of California charged three attorneys over AI-fabricated citations in April, and five months later no public resolution has been reported. Court sanctions punish the filing. Bar discipline, when it lands, punishes the lawyer.
The Idaho Supreme Court used the strongest language of any ruling this week and imposed the least. That is less a contradiction than a reminder of what a warning is for: it puts the next litigant, in that court, on notice, without resolving anything for this one. A licensed lawyer reading Posey v. Bushnell should not take comfort in the lack of sanction; Bushnell had no bar card for the court to threaten. The two pending federal show-cause orders, naming lawyers instead, are the ones worth watching for what a comparable warning costs when a license is on the table.
California's proposed AI-specific amendments, which would require lawyers to independently verify every AI output, closed public comment on May 4 with no adoption date announced since. Malpractice insurers did not wait. The five requirements described this week, a written policy, independent verification, controlled data environments, training, and a clear owner of the judgment, are not law anywhere yet. For any firm that wants coverage, they are close enough.
Exhibit AI
On September 18, a judge in the Eastern District of Pennsylvania entered a show-cause order in Whitfield v. Prime Agency after finding two case citations that do not exist, three quotations attributed to real cases that do not contain them, and three additional holdings the court found misrepresented. Eight errors, one filing, one signature. The order requires a response before the court decides whether sanctions follow; none has been made public yet.
Nothing about this pattern is unusual. It is the same failure mode this publication has covered for weeks: a citation check that either did not happen or did not catch what it was meant to catch, in a filing that carried a real signature. What makes Whitfield worth sitting with is timing. It landed the same week a state supreme court declined to sanction a nearly identical failure, because that filer had no license to protect. This one does.
The Lesson
Eight errors in one filing is not a single missed citation; it is a verification step that was skipped entirely, not merely done badly. Build the check into the filing workflow itself, as a step nobody can sign around, rather than trusting any one reviewer to remember to run it.