■ SAFE RISK ■ Legal
AI will draft the routine orders and search the precedents, but judging is the state exercising legitimate authority over human lives — and societies show near-zero appetite for handing sentencing to software. The bench stays human; the chambers get automated.
“AI can analyze precedents, but 'All rise for the algorithm' doesn't inspire confidence.”
Our AI replacement risk score — how we score jobs
Judges run courtrooms and decide what the law means when people disagree about it. The workload mixes trial management — ruling on objections in real time, managing juries, keeping lawyers in line — with an enormous paper practice: reading briefs, researching precedent, writing opinions and orders, and clearing dockets clogged with routine motions, pleas, and scheduling disputes. Sentencing, custody decisions, and bail determinations put a human being's liberty or family on the line, often based on credibility judgments about who is lying and who is scared.
The chambers side is automating quickly. Legal research tools with AI summarization compress days of precedent-hunting into hours; draft orders and routine opinions can be machine-generated for judicial editing; and some jurisdictions already use algorithmic risk-assessment tools in bail and sentencing support — controversially, since bias in those systems has drawn years of well-earned criticism. A few countries have experimented with automated adjudication of small claims and traffic matters, and online dispute resolution keeps absorbing low-stakes conflicts entirely. Court backlogs make this attractive: the pressure to automate routine adjudication is fiscal, not just technical.
The core resists for reasons deeper than capability. Judicial legitimacy is the product — rulings bind because a sworn, accountable human made them through visible process, and appellate courts exist precisely to review human reasoning. Credibility assessment in live testimony, mercy and proportionality in sentencing, and novel questions where precedent runs out all require judgment the public insists stays human; surveys and legal-profession bodies consistently reject machine adjudication of serious matters. Constitutionally, many systems simply cannot delegate judicial power to software. Expect judges with AI-drafted paperwork and machine-triaged dockets, fewer clerkships doing grunt research, and the robe itself untouched — the risk score here reflects transformed workflow, not endangered judges.
Automatability: our editorial assessment of current and near-term AI capability
Chambers work is transforming now — AI research and drafting assistance is entering courts this decade, and small-claims and traffic matters will increasingly resolve through automated or online processes by 2030. Judgeships themselves are constitutionally insulated; no serious jurisdiction plans machine adjudication of criminal or significant civil matters by 2040. The downstream effect lands on clerks and staff attorneys, whose research hours are the actual automation target.
For parking tickets, effectively yes — automated and online dispute resolution already handles low-stakes matters in several jurisdictions. For anything touching liberty, family, or serious money, no: judicial power is constitutionally vested in humans in most legal systems, and public trust in machine adjudication is minimal. The realistic future is AI-assisted judges, not artificial ones. Our risk score of 23 is about the paperwork, not the robe.
Mostly behind the bench: AI-powered legal research and summarization, draft generation for routine orders, docket triage, transcription, and — most controversially — algorithmic risk scores informing bail and sentencing decisions, which have drawn sustained criticism over bias. Courts have also had to police AI misuse by lawyers, sanctioning filings with hallucinated citations. Adoption is uneven and deliberately cautious; legitimacy is the constraint.
Yes — judgeships are among the most automation-insulated positions in law, protected by constitutional design, life tenure in some systems, and public expectation. The route to the bench is changing more than the bench itself: the associate and clerk work that traditionally built judicial résumés is exactly what AI is absorbing, so aspiring judges should build courtroom, trial, and judgment-heavy experience rather than research-heavy credentials.
Genuinely contested. Consistency tools could reduce the documented arbitrariness in human sentencing — outcomes shouldn't depend on the judge's lunch schedule — but risk-assessment algorithms trained on historical data have reproduced and laundered old biases behind a veneer of objectivity. The emerging consensus: AI as a transparency and research aid, with humans owning discretionary calls and every algorithmic input open to challenge in court.