Loading today’s stories
Build1 distinct publisher3 min readUpdated
A Georgia chief justice counted the fabrications aloud from the bench. Nine days later the district attorney had filed a bar grievance against her own appellate prosecutor.
The Engineer · Build desk
Compiled by The EngineerSomething wrong?How this is made
On 18 March 2026, an assistant district attorney stood at the lectern of the Supreme Court of Georgia in downtown Atlanta and tried to explain why several of the cases in her brief did not exist [1]. That hearing is the cleanest artefact available for anyone who has been asked to justify a verification step: a named professional, an itemised count from the bench, and a documented chain of licensing and employment consequences inside nine days [6][12][13]. Deborah Leslie was an assistant district attorney in Clayton County, assigned to appellate work and assets forfeiture [2]. The filing at issue opposed a new trial for Hannah Payne, convicted in 2023 of the murder of Kenneth Herring after a 2019 hit-and-run and serving life with the possibility of parole [3]. Payne's lawyer, Brian Steel, had argued her original counsel failed to ask the jury to consider citizen's arrest as a defence [4]. The state's response, signed by Leslie, ran to dozens of pages [5]. Chief Justice Nels Peterson counted the defects aloud: at least five citations to cases that did not exist, and at least five more to cases that existed but did not say what the brief claimed [6]. That is at least ten bad citations in a single document [7]. Clips of the exchange were later viewed more than five million times [8]. Leslie's first explanation was that the citations might have been added to the version filed with the court rather than the one she drafted [9]; Peterson observed that the same non-existent cases appeared in the brief opposing Payne's motion in the court below [10]. The employer did not wait. On 27 March, Clayton County District Attorney Tasha Mosley wrote to the Chief Justice, in a letter published shortly afterwards by local outlets, conceding that Leslie had used artificial intelligence to draft the filing and had not verified the output [11]. Nine days elapsed between the hearing and the letter [12]. The office filed a grievance with the State Bar of Georgia and imposed suspension, a performance improvement plan and loss of privileges [13]. The operative sentence is Leslie's own. In her affidavit she said the errors were not intentional and that the references "were not independently verified before inclusion" [14]. No finding of bad faith is required for the professional consequences to land; the absence of a check was sufficient [13][14]. None of this is isolated. Damien Charlotin, who divides his time between Sciences Po Law School and HEC Paris, began a database of these incidents in April 2025 because he could not find anyone else doing it [15]. By spring 2026 the AI Hallucination Cases tracker had passed 1,200 documented incidents worldwide, roughly 800 of them from the United States, which is about two thirds of the total [16][19]. On one day in March 2026 he logged seventeen [17], several times the average daily rate implied by the tracker's own totals [20]. Charlotin says the rate is still rising [18]. The demand pressure behind that curve is not going away. The Legal Services Corporation's 2022 Justice Gap study with NORC at the University of Chicago found that 92 per cent of substantial civil legal problems of low-income Americans receive no or insufficient legal help, and that 74 per cent of low-income households face at least one such problem a year [21][22]. Ministry of Justice figures for the third quarter of 2025 showed 59 per cent of County Court civil cases in England and Wales had at least one unrepresented party [23]. Self-representation routinely exceeds 90 per cent in US state housing, family and consumer dockets [24]. Watch the State Bar of Georgia's disposition of the grievance, since that is where the personal cost gets priced. Watch what happens to the Payne appeal, which the dev.to account describes as compromised by the filing [25].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
On the morning of 18 March 2026, Deborah Leslie stood at the lectern of the Supreme Court of Georgia in downtown Atlanta and tried to explain why several of the cases in her brief did not exist.
The state's response, signed by Leslie, ran to dozens of pages and cited authorities that in many places were imaginary.
From the bench, Chief Justice Nels Peterson counted aloud at least five citations to cases that did not exist, and at least five more to cases that existed but did not say what Leslie's brief claimed they said.
Leslie was an Assistant District Attorney with Clayton County, assigned to appellate work and assets forfeiture.
Leslie had filed papers opposing a new trial for Hannah Payne, convicted in 2023 of the murder of Kenneth Herring after a hit-and-run on a Clayton County road in 2019; Payne, then twenty-five, was serving life with the possibility of parole.
Payne's lawyer, Brian Steel, filed for a fresh trial on the grounds that her original counsel had failed to ask the jury to consider citizen's arrest as a defence.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Specific and checkable, but single-publisher and entirely second-hand
The narrative core is unusually concrete for a single source: named court and chief justice, a dated hearing, a counted tally of defective citations, a dated district attorney letter published by local outlets, quoted affidavit language, and named statistical studies (LSC/NORC 2022, MoJ Q3 2025). Against that, nothing primary is supplied — no docket number, brief, letter text or bar filing — there is no second publisher in the cluster, and two supporting assertions (the five-million video views and the 90-per-cent state-docket self-representation rate) are unsourced generalisations.
AI use in court filings is documented at institutional scale
Adoption here is real-world use of generative AI in legal drafting, and it is evidenced rather than projected: one prosecutor's office admits an AI-drafted, unverified appellate brief and takes disciplinary action, while an independent academic tracker records more than 1,200 court incidents worldwide since April 2025, roughly 800 in the United States, including seventeen logged on a single March 2026 day. The ceiling is set by what is absent — no tool or vendor is named, no adoption denominator (filings, firms, offices) is given, and no data show how many courts have adopted verification or disclosure rules in response.
Documented core, essayistic framing slightly ahead of the evidence
The factual spine — counted fabrications, an admitted AI draft, a nine-day escalation to a bar grievance, a 1,200-incident tracker — is reported with dates and named actors, so the story is not inflated at its centre. The overshoot is at the edges: the title and dek dramatise a single-jurisdiction disciplinary event into a general policy-deck exhibit, the appeal is described as 'compromised' with no reported disposition of Payne's motion, the rising-rate claim rests on one researcher's assertion without a time series, and the five-million-views and 90-per-cent self-representation figures are unsourced.
Interested parties shape the causal account of a documented event
The two most load-bearing attributions come from parties with clear stakes. The district attorney's office both conceded the AI use and simultaneously sanctioned the individual prosecutor, an account that locates fault in one employee rather than in office supervision or filing controls; Leslie's affidavit, filed while facing a bar grievance and suspension, stresses lack of intent. The publisher itself shows no visible product or vendor interest — no tool is named or promoted — and the tracker is maintained by an academic with no disclosed commercial stake, which keeps the score mid-range rather than high.
Confident on the incident, thin on generalisation and outcomes
Confidence is high for the dated, countable specifics of the Georgia matter and for the existence and scale of the hallucination tracker, all of which are internally consistent and readily checkable against public court and academic records. It is materially lower for the broader argument, because the cluster has one publisher, no primary documents, no outcome for either the Payne motion or the bar grievance, and two unsourced quantitative claims propping up the access-to-justice framing.
science
Text watermarks land on 2 December. The detection they imply does not.1 distinct publisher
invest
A Connecticut judge just priced prompt injection: no fine, no e-filing2 distinct publishers
science
Claude's watermark is a compliance artefact, not a cheating detector1 distinct publisher
security
OpenAI's 13-17 tier turns teen AI safety into an age-assurance problem1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
dev.to
1 article · August 16, 2026