A Nevada judge was accused of letting AI write her courtroom ruling. A federal court said: even if that’s true, she can’t be sued for it. Meanwhile, 60% of U.S. federal judges are now using AI tools on the job. The question is no longer whether AI will enter the courtroom. It’s already there. The question is how far we let it go.
Should AI decide your court case? Currently, no major legal system permits AI to make final judicial decisions autonomously. However, AI is actively used in U.S. courts for risk assessment, sentencing guidance, and legal research and judges are already using AI to draft rulings. A 2026 federal court ruling confirmed that judges are protected by judicial immunity even if they rely “wholly on artificial intelligence” to decide a case. The debate is no longer theoretical.
Why White-Collar Workers Are Leaving Their Jobs Before AI Replaces Them
Most people imagine AI in courts as a science fiction scenario, a robot judge banging a digital gavel. The reality in 2026 is far quieter and, in some ways, more alarming. AI has not entered the courtroom with drama. It has slipped in through the back door, in the form of risk assessment tools, AI-assisted legal research platforms, and most recently AI drafting of judicial opinions themselves.
According to a study released by Reuters in March 2026, 60% of U.S. federal judges now report using at least one generative AI tool in their judicial work. The tools being used include ChatGPT, Claude, CoCounsel, Westlaw AI-Assisted, Deep Research, and Lexis+ AI. These are not fringe experiments. They are mainstream tools in active daily use across federal courtrooms.
Furthermore, AI risk assessment tools such as the COMPAS algorithm and the Public Safety Assessment (PSA) have been used in American courts for years to inform bail decisions, sentencing, and parole recommendations. In those contexts, AI is not merely assisting the judge. It is directly shaping the outcome of real cases affecting real people’s freedom.
The Difference Between Assisting and Deciding
There is a critical distinction that gets lost in most discussions about AI in courts: the difference between assisting a judge and deciding for a judge. These two things sound related. In practice, they sit on opposite ends of an ethical cliff.
An AI that helps a judge quickly research precedent from 10,000 cases is an efficiency tool. An AI that tells a judge “this defendant has a 78% probability of reoffending” and where the judge treats that number as a determining factor rather than one data point among many is something entirely different. That is algorithmic decision-making with a human face put over it.
The Nevada Case That Changed the Conversation About AI and Judicial Power
Phillips v. Parlade — Federal Court Rules Judicial Immunity Covers AI-Written Rulings
In August 2026, a federal court in Nevada was asked to rule on one of the most provocative allegations in recent legal history: that a state court judge had farmed out her judicial opinion entirely to artificial intelligence.
The plaintiff, who had appeared before Nevada family court judge Mari Parladé, accused her of relying wholly on AI to issue her ruling. His evidence: the judge appeared to read from a prepared script during the hearing, stumbled over a word mid-sentence, and accidentally said “appearance of any impartiality” instead of “appearance of impropriety” suggesting she was reading AI-generated text she had not fully reviewed.
Federal Judge Gloria Navarro dismissed the lawsuit. Her reasoning was blunt: even if Judge Parladé had relied entirely on artificial intelligence to write her ruling, issuing a judicial ruling is a normal judicial function. Therefore, judicial immunity applies. The plaintiff cannot sue a judge in federal court for how she reasons or what she uses to reason with.
Legal commentator Eugene Volokh of the Volokh Conspiracy was careful to note that the judge probably did not actually let AI decide the case. Nevertheless, the ruling’s legal significance stands independently of what actually happened. As a matter of federal law, a judge who relies wholly on AI is still immune from civil lawsuits over that decision.
Consequently, state-level remedies appeals, disciplinary proceedings, mandamus petitions remain available. However, the practical takeaway is deeply uncomfortable: a judge can use AI to decide a case, and the affected party may have limited federal recourse.
COMPAS and the Bias Problem That AI in Courts Has Never Solved
Before the Nevada case grabbed headlines, the most important story about AI in courts was quieter — and in many ways worse. It is the story of the COMPAS algorithm.
What Is COMPAS?
COMPAS — Correctional Offender Management Profiling for Alternative Sanctions — is a risk assessment tool used in U.S. courts to predict a defendant’s likelihood of reoffending. Judges use COMPAS scores to inform bail and sentencing decisions. The algorithm asks defendants 137 questions. It then produces a risk score from 1 to 10. The higher the score, the more likely, according to the model, that the defendant will commit another crime.
On its surface, COMPAS sounds rational. It sounds scientific. It sounds like it removes the messy inconsistency of human judgment from a process where human inconsistency has historically produced wildly different sentences for identical crimes depending on which judge a defendant drew.
What ProPublica Found — And Why It Matters in 2026
In 2016, ProPublica published an investigation that shook the criminal justice world. Their analysis found that COMPAS incorrectly flagged Black defendants as future criminals at nearly twice the rate of white defendants. Specifically, Black defendants who did not reoffend were nearly twice as likely to be labeled high-risk compared to white defendants who also did not reoffend.
Northpointe, the company behind COMPAS, disputed the methodology. Academic researchers debated the statistical framing. However, the core problem proved durable: an AI trained on historical criminal justice data will learn and then amplify the biases embedded in that history. The system learns that certain zip codes, certain demographics, and certain socioeconomic backgrounds correlate with reoffending. Then it uses those correlations to predict the future. And the future, according to the algorithm, looks suspiciously like the past.
A 2026 academic review published on arXiv confirmed that the underlying problem has not been resolved. Despite years of debate, empirical evidence shows that AI tools used in pretrial and sentencing contexts have modest real-world impact but carry significant risks of encoding and perpetuating systemic inequality.
“An AI trained on biased historical data does not produce neutral decisions. It produces systematized bias — bias with the appearance of objectivity.”
When a Judge Let AI Draft the Ruling — And an Appeals Court Called It a “Botch Job”
The Nevada case involved allegations. The Mississippi case involved a confession.
US District Judge Henry Wingate acknowledged in 2025 that his law clerk had used the AI tool Perplexity to help draft an order in a case challenging a Mississippi law banning the use of public education funds on diversity, equity, and inclusion programs. The resulting order which blocked the Mississippi law was riddled with AI-generated hallucinations: references to cases that did not exist, misstatements of legal precedent, and errors of law.
In October 2026, a three-judge panel of the U.S. Court of Appeals for the Fifth Circuit vacated the injunction and ordered the case reassigned to a different judge. The panel’s language was unusually harsh. Judge Jerry Smith, writing for the appellate court, described the situation as “a botch job.” He noted that the errors of law embedded in the AI-hallucinated order made it impossible for the original judge to fairly reconsider the case.
5th Circuit Reassigns Mississippi DEI Case After AI Hallucinations in Ruling
Judge Henry Wingate’s law clerk used Perplexity AI to help draft a court order. The resulting opinion contained fabricated case citations and errors of law. The 5th Circuit vacated the ruling, called it “a botch job,” and reassigned the case to a new judge.
The incident illustrates the specific danger of AI hallucinations in legal contexts. In a general setting, a hallucinated fact is an error. In a court ruling, a hallucinated case citation is a legal fiction and a defendant’s freedom may rest on it.
How Different Countries Are Handling AI in Courts — A Global Comparison
The question of AI in courts is not a uniquely American problem. Every major legal system is grappling with it and they are arriving at very different answers.
| COUNTRY | AI DECISION-MAKING | AI ADVISORY USE | NOTABLE DEVELOPMENT |
|---|---|---|---|
| United States | Limited | Active (60% of judges) | COMPAS in sentencing; judicial immunity covers AI rulings (2026) |
| China | Yes — minor cases | Active | Smart courts with AI judges handling civil cases |
| India | Banned | Proposed | Supreme Court 2026 draft bans AI from sentencing and bail |
| Estonia | Pilot only | Active | AI decides small claims under €7,000 (~$7,700) |
| European Union | Restricted | Regulated | EU AI Act classifies justice AI as “high-risk” |
| UAE | Commercial courts only | Active | AI-assisted adjudication tested in commercial dispute resolution |
| UK | No | Growing | Courts and Tribunals Service piloting AI research tools for judges |
China’s Smart Courts — The World’s Most Advanced AI Justice System
China is the only major nation that has actually deployed AI judges in an operational capacity. Through a network of “smart courts,” AI systems handle certain minor civil cases particularly small contract disputes and traffic violations with minimal human judicial involvement. The system has processed millions of cases.
Proponents argue that smart courts dramatically reduce case backlogs and deliver consistent, fast justice for low-stakes disputes. Critics point out that China’s smart courts operate without the due process protections that define Western legal systems — no independent appeals process, no transparent algorithmic accountability, and no adversarial challenge to the AI’s reasoning.
India’s Supreme Court Takes the Opposite Position
In 2026, India’s Supreme Court released draft regulations on AI in courts that take one of the strongest positions against autonomous AI decision-making of any legal system in the world. The draft completely prohibits AI from determining judicial outcomes, delivering sentences, or deciding bail eligibility. It establishes a dedicated Apex Body and Centre of Research and Excellence on Artificial Intelligence (CoRE-AI) to regulate nationwide adoption. Human judicial primacy is described as absolute and non-negotiable.
What Should Actually Happen with AI in Courts
The question “should AI decide your court case?” has a clear answer right now: No. Not because AI cannot be useful in legal contexts. It clearly can be. But because the current state of AI — hallucination-prone, bias-carrying, opacity-generating — makes it unsuitable for final judicial decision-making in any domain where human freedom is at stake.
However, the question will not stay this simple forever. As AI systems improve, as their outputs become more auditable, and as legal frameworks for algorithmic accountability mature, the line between “assisting” and “deciding” will blur further. The Nevada immunity ruling suggests the law is not yet keeping pace with the technology.
What Responsible AI Use in Courts Looks Like
- AI for research, not reasoning. AI can legitimately help judges and clerks surface relevant case law faster. It should not replace the reasoning process itself.
- Mandatory disclosure. If AI tools influenced a ruling, defendants should have the right to know and to challenge the AI’s output.
- Auditable algorithms. Any AI used in sentencing or bail decisions should be subject to independent audit for bias and accuracy.
- Human final authority. Every consequential judicial decision must be made and owned by a human judge who can be held accountable.
- Sunset clauses on risk tools. AI risk assessment tools should be regularly re-evaluated rather than allowed to embed themselves permanently into sentencing practice.