AI jury selection tools now scan a prospective juror’s social media history, public records, and voting patterns within minutes of a name appearing on a jury list, and the American Bar Association’s own ethics guidance on the practice, issued just last year, still leaves the hardest question unanswered. Formal Opinion 517, published July 9, 2025, tells lawyers they cannot follow an AI program’s strike recommendation if they know or should know it amounts to unlawful discrimination. What it does not tell them is how a lawyer is supposed to know that, when the software making the recommendation is proprietary and its underlying methodology is not disclosed.
That gap sits at the centre of a technology that has moved from a novelty used in a handful of high-stakes trials to a routine part of jury selection at firms that can afford it. This piece looks at what AI jury selection tools actually do, what the legal profession’s own ethics body has said about their limits, and why the deeper problem the technology touches predates AI by more than a century.
What AI Jury Selection Tools Actually Do
The current generation of jury selection software falls into two broad categories. The first uses web-scraping bots and natural language processing to build a profile of each prospective juror from public sources: social media posts, voting records, property records, and news mentions, compressed into a report a lawyer could never assemble by hand in the narrow window voir dire typically allows. Companies including Magna Legal Services and Vijilent offer this kind of research-and-surveillance product, built to surface a juror’s public digital footprint before a lawyer ever asks a question in court.
The second category goes further, applying a scoring algorithm to rank prospective jurors by how favourable they are likely to be to a given side. Momus Analytics and similar platforms feed juror questionnaire responses, demographic data, and case-specific attributes into a model trained on historical verdict patterns, then output a ranked list with peremptory strike recommendations. Richard Gabriel, president of the trial consulting firm Decision Analysis, told the ABA Journal that AI is genuinely useful for gathering data but becomes questionable the moment it moves into interpreting or scoring a juror’s likely leanings, since qualities like leadership or persuadability in a jury room are not reliably visible in someone’s public employment history or social media activity.
The Ethics Opinion That Raised More Questions Than It Answered
The ABA’s Formal Opinion 517 is the most direct statement yet from the legal profession on where AI jury selection crosses an ethical line. The opinion, addressing Model Rule of Professional Conduct 8.4(g), states plainly that a lawyer who knows or reasonably should know that exercising a peremptory challenge constitutes unlawful discrimination violates professional conduct rules, regardless of whether the recommendation came from a client, a jury consultant, or an AI program. The opinion specifically addresses the scenario where AI software ranks jurors and applies those rankings in a discriminatory manner without the lawyer’s direct knowledge.
The standard the opinion sets is due diligence: lawyers must acquire, in the ABA’s words, a general understanding of the methodology a juror selection programme uses before relying on its output. In practice, that requirement runs into a wall most litigators are not equipped to climb. The scoring algorithms behind commercial jury selection software are proprietary, and vendors have limited commercial incentive to open their models to the kind of technical scrutiny that would let a lawyer verify a tool is not using race, gender, or another protected characteristic as a hidden proxy variable, buried inside factors like neighbourhood, employer type, or media consumption habits that correlate with those characteristics without naming them directly.
The Problem AI Inherited, Not the One It Created
It would be a mistake to treat AI jury selection as introducing bias into a previously neutral process. The Equal Justice Initiative’s research on jury selection documents that racial discrimination has persisted at every stage of the process for nearly 150 years since Congress first outlawed it, from how jury pools are compiled to how peremptory strikes are actually used, and that courts, prosecutors, and legislatures have largely tolerated the pattern rather than enforced against it. Several states have implemented reforms or launched formal studies of the problem; EJI notes that most states have done essentially nothing.
What AI changes is not whether bias exists in jury selection but how it is laundered. A lawyer who strikes jurors along racial lines using their own judgement leaves a pattern a defence attorney can potentially challenge under Batson v. Kentucky, the 1986 Supreme Court decision requiring a race-neutral explanation for a suspicious pattern of strikes. A lawyer who strikes the same jurors based on an opaque algorithm’s composite score has a ready-made, technically plausible explanation available, one a court has no practical way to interrogate without access to a vendor’s proprietary model. The technology does not need to be built with discriminatory intent to make discriminatory outcomes considerably harder to prove.
The Vendor’s Case, and Its Limits
Jury selection software vendors do not market their products as discrimination tools, and the more credible ones make a point of building in restrictions against it. Momus Analytics states explicitly that its scoring system does not use protected characteristics like race, sex, or national origin as inputs. That claim is worth taking seriously rather than dismissing outright, since the alternative to AI-assisted jury selection is not a bias-free process; it is the same human intuition and pattern-matching that produced 150 years of documented discrimination in the first place.
The harder problem is proxy discrimination, where a model never uses a protected characteristic directly but instead relies on variables so tightly correlated with it, such as neighbourhood, employer type, or specific media consumption habits, that the practical effect is functionally identical to the discrimination the model claims to avoid. A vendor’s assurance that race is not an input does not resolve whether the ten variables the model does use are, in combination, doing the same work. Verifying that distinction requires exactly the kind of technical access to a model’s internals that ABA Formal Opinion 517’s due diligence standard calls for and that the current commercial market rarely provides.
What This Means for a Defendant
For someone facing trial, the practical consequence of AI jury selection is asymmetric access. High-value commercial litigation and death penalty defence work, the cases with budgets to afford sophisticated jury consulting, are where these tools appear most often.
A criminal defendant relying on a public defender with no budget for jury consulting software faces a prosecution that may have full access to a scored, data-driven strike strategy, while the defence is working with instinct, experience, and a fraction of the preparation time. That resource gap predates AI, since traditional jury consulting has always been expensive, but AI tools have lowered the cost of the data-gathering component enough to make sophisticated jury profiling accessible to a wider range of well-funded litigants without doing anything to close the gap for those who are not.
The same asymmetry shows up in a different form in how sentencing decisions get made after a trial concludes. LiveAIWire’s earlier coverage of AI sentencing bias in predictive risk tools found a nearly identical structural pattern: a proprietary algorithm influencing a consequential decision about a person’s liberty, deployed with minimal independent audit, and defended largely on the basis of vendor-commissioned validation rather than transparent, court-accessible methodology.
Jury selection and sentencing sit at opposite ends of the same trial, and AI has entered both through the same door: a genuine efficiency and information advantage for whichever side can afford it, layered on top of a justice system that has never fully solved the discrimination problem the technology now has new tools to obscure.
Where the Accountability Gap Actually Sits
The due diligence standard ABA Formal Opinion 517 sets is a meaningful step, but it places the burden entirely on individual lawyers to interrogate technology that vendors have no obligation to make interrogable. That mismatch is not unique to jury selection. LiveAIWire’s coverage of facial recognition and algorithmic tools used to police the police documented the same underlying failure mode: courts and oversight bodies repeatedly asked to evaluate the fairness of a proprietary system without being given the technical access needed to actually do so, leaving verification to rest on the vendor’s own assurances.
The broader accountability infrastructure needed to close that gap, tools and legal frameworks that let an independent party audit an algorithm used against the public without requiring the vendor’s cooperation, remains underdeveloped relative to how quickly these tools have spread through the justice system. LiveAIWire’s reporting on the growing movement to build that kind of independent audit capacity found the same pattern recurring across sector after sector: the technology arrives, gets deployed, and only afterward does anyone build the capacity to check whether it is doing what its vendor claims.
What This Means for You
If you are ever seated in a jury pool, the honest reality is that some version of your public digital footprint, social media history, voting record, property records, may already have been reviewed by an algorithm before you are asked a single question in court, and you will likely never know which side reviewed it, what it concluded, or whether that conclusion factored into whether you were struck. There is currently no requirement in most jurisdictions that AI-assisted jury selection be disclosed to prospective jurors, and no established mechanism for a struck juror to challenge the basis of that decision the way a criminal defendant can challenge a discriminatory pattern under Batson.
For anyone involved in a case, criminal or civil, where the opposing side may be using sophisticated jury consulting technology, the ABA’s own guidance suggests the relevant question to ask is not whether AI was used but whether anyone, including the lawyer relying on its output, actually understands how the tool arrived at its recommendations. Given how rarely that understanding is possible with current commercial products, the honest answer in most cases is likely to be no.
The Unresolved Question at the Centre of This Technology
AI jury selection sits at the intersection of two problems the legal system has never fully solved on its own: a jury selection process with a well-documented, nearly 150-year history of racial discrimination that formal doctrine has only partially curbed, and a legal profession’s ethics rules that were not written with proprietary black-box software in mind. The ABA’s Formal Opinion 517 is an honest attempt to extend existing anti-discrimination doctrine to a new technology, but an ethics opinion that tells lawyers to understand a methodology vendors have no obligation to disclose is, in practice, asking for a level of transparency the current market for this software simply does not provide.
Until that changes, whether AI makes jury selection fairer or simply makes existing unfairness harder to detect will depend less on the sophistication of the algorithm than on whether anyone outside the vendor’s own walls is ever allowed to actually look inside it.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity, and the social impact of emerging technology. He publishes daily at LiveAIWire.com.