An associate at a mid-size law firm has a brief due at 5 PM. She asks an AI tool to find supporting case law, copies three citations into her filing, and sends it to the partner for sign-off.
The partner, busy, signs.
Two of the cases don’t exist, and opposing counsel notices. The judge asks both sides to produce the cited opinions, and by Friday the law firm is explaining itself in open court.
That scene is no longer rare. 1,621 court cases worldwide now involve generative AI producing hallucinated content in legal filings, with the count climbing daily.
On June 8, 2026, a federal judge in Mississippi removed every lawyer from a case after finding AI-fabricated sources in filings from both sides.
Burnout is a leading cause of these outcomes, with attorneys reported feeling burned out 42% of the time, according to Bloomberg Law.
While this drives practitioners toward AI, it also calls for caution in the legal industry. The same technology that speeds up legal work can sink it.
Why legal work suits AI, and why it’s exposed
Legal work runs on text, patterns, and structure. Those are the conditions where generative AI performs best, which is why adoption among legal professionals has moved fast.
69% of legal professionals now use general-purpose AI for work tasks, up from 31% a year earlier.
Think AI legal research or putting together a professional services report.
Firm-wide adoption tells a slower story. Only 34% of firms have implemented legal-specific AI tools, against 46% using general-purpose platforms.
That divide—individual lawyers experimenting ahead of any firm policy—is where most of the risk concentrates.
The legal workflows AI handles well
Generative AI doesn’t improve every legal task equally. The workflows it excels at share two traits: high volume and low ambiguity.
The table below maps seven of these legal ops use cases, with the level of oversight each one needs.
| Practical law use case | What AI does well | Human oversight |
| Legal research and case law review | Surfaces relevant precedent and summarises long opinions in minutes | Active verification of every citation |
| Contract review and contract analysis | Flags missing clauses, inconsistencies, and risk terms across long documents | Routine review |
| Legal drafting and contract drafting | Produces first drafts of NDAs, letters, and standard agreements | Routine review |
| Evidence review and chronologies | Sorts high-volume files and builds timelines | Active verification |
| Client intake and AI matter management | Collects structured information and routes new cases | Routine review |
| Billing, time-tracking, and admin | Drafts time entries and catches invoice errors | Routine review |
| File QA, compliance, and document automation | Checks a legal document for consistency and standard compliance at scale | Routine review |
Across these workflows, the potential productivity gains are huge. Thomson Reuters estimates AI could free roughly 240 hours per legal professional each year.
Lawyers using AI completed 12% more tasks and finished 25% faster than those not using AI, according to a legal trends report by Clio that surveyed 1,702 U.S. legal professionals.
Vendors now package several tasks into a single AI legal assistant. CoCounsel Legal, from Thomson Reuters, combines research, review, and drafting.
Some legal teams extend this to AI contract negotiation, where the model proposes redlines against a playbook.
The condition attached to each use case is that a human must check the output before it reaches a client or court.
The work legal AI tools shouldn’t touch
A second category of legal work resists automation, because the value lies in judgment, not output.
AI shouldn’t own any of the following:
- Legal advice and risk assessment
- Litigation strategy
- Settlement decisions and the final terms in any AI contract negotiation
- Ethical calls and the duty of candor to a court
- Final sign-off on any court filing
The main reason is hallucination. Stanford’s RegLab tested the leading legal research platforms and found that they hallucinate between 17% and 33% of the time.
This happened despite retrieval systems built to prevent such hallucination. General-purpose models score worse.
The ABA’s Formal Opinion 512 requires lawyers and legal departments to keep a reasonable understanding of what AI can and can’t do.
Competence includes knowing where AI algorithms break. No legal AI tool removes a lawyer’s duty to verify.
How to map your firm’s legal AI workflow readiness
Before scaling any AI automation, legal teams can score each task on two axes: impact and risk.
Impact is the time saved and quality gained. Risk covers data sensitivity, ethical exposure, and how visible the work is to clients.
Automate tasks that score high on impact and low on risk. This usually means internal legal ops, not sensitive cases.
Bring partners, associates, support staff, and IT into the decision, because each sees a different part of the legal workflow.
| Readiness tier | Marker | Typical next step |
| Experimenting | Individual lawyers use AI tools with no firm policy | Write an AI use policy and approve specific tools |
| Enabled | The firm has approved tools and basic rules | Add citation verification protocols and audit trails |
| Embedded | AI runs inside legal workflow automation with oversight | Extend to agentic AI under human review |
Most firms today fall into the first two tiers: Experimenting and Enabled.
The move to Embedded, where automation and oversight operate together across legal operations, is the stage governance frameworks now aim for.
The governance layer every legal team needs
AI governance is no longer optional, but enforcement has been mixed globally.
In the US, since the first federal AI standing order in 2023, hundreds of state and federal judges have amended or issued standing orders, local rules, and pretrial orders on AI use and verification.
In the UK, the Courts and Tribunals Judiciary updated its AI guidance for judicial office holders in October 2025, requiring personal responsibility for all AI-assisted output. No express disclosure mandate for lawyers exists yet.
The Law Society called on the SRA and MoJ in May 2026 to issue binding rules. The Civil Justice Council is consulting. Guidance exists, but enforcement is still forming.
In APAC, Singapore leads. Its Ministry of Law published a Guide for Using Generative AI in the Legal Sector in March 2026, and all three court systems issued AI circulars in September 2024.
Australia has issued court-specific AI practice notes across multiple jurisdictions, including the Federal Court and the Supreme Courts of NSW, Victoria, and Queensland.
South Korea’s judiciary published its own AI guidelines in February 2025.
In Africa, courts have moved faster than legislators. South African courts found fabricated AI citations in filings and ruled lawyers personally liable.
A Kenyan Supreme Court judge publicly warned practitioners against using generative AI in proceedings after fabricated authorities appeared on the record in 2025.
Bar associations, international courts, and state statutes across jurisdictions all converge on the same three duties: document the tools, verify the output, and disclose the use.
Legal output generated by AI technology must be able to withstand scrutiny if challenged in court.
I help teams map their governance posture, manage AI workflows, and implement fiduciary grade AI in their legal practice. Book an audit today.