AI Lawyer Hallucinations: Real Court Cases and Sanctions From 2025–2026

When lawyers ask whether they can use ChatGPT for legal work, the safer question to ask is: would I be willing to defend my use of it before a federal judge? Because by mid-2026, that’s exactly what’s happening with increasing frequency. Courts have moved from issuing cautionary opinions to handing down five- and six-figure sanctions, and at least 35 state bar associations now require attorneys to verify AI-generated content.

This article catalogs the major hallucination cases and sanctions from 2025–2026, what attorneys actually did wrong, and what the current state of the rules is.

The Scale of the Problem

Researcher Damien Charlotin has been maintaining a public database of AI hallucination cases — instances where attorneys filed AI-generated work containing fabricated case citations, invented quotations, or non-existent authorities. As of mid-2026, the database has cataloged over 1,353 incidents globally, with the pace accelerating sharply in early 2026.

Most of these don’t make headlines. They’re routine sanctions: $1,000 here, $2,500 there, a Rule 11 referral, a state bar complaint. But the headline cases tell you where this is heading.

Major Sanctions in 2026

Oregon Federal Court — $110,000 (the record)

A federal judge in Oregon sanctioned two lawyers a combined $110,000 in 2026 — the largest AI hallucination penalty in American legal history. The attorneys had submitted a brief containing 23 fabricated citations and 8 invented quotations. The court reviewed the filing, found that the cases and quotes didn’t exist, and concluded that the attorneys had failed in their basic duty of verification.

Sixth Circuit (Cincinnati) — $30,000

The 6th Circuit imposed a $30,000 combined sanction against two attorneys for submitting filings with more than two dozen fake case citations. The court’s order emphasized that an attorney’s signature on a brief is itself a representation that the cited authorities exist and stand for what the brief says they stand for.

Southern District of Ohio — $7,500

Senior Judge Walter H. Rice of the U.S. District Court for the Southern District of Ohio imposed a collective $7,500 sanction against two attorneys for AI hallucinations in court documents. Smaller in dollar terms, but representative of dozens of similar orders being entered across federal courts in 2026.

Cherry Hill, New Jersey — Repeat Offender

A federal judge sanctioned a Cherry Hill, NJ attorney for filing a brief with AI hallucinations — and notably, it was the second sanction against the same attorney for the same conduct. Courts are now treating repeat AI-hallucination conduct as significantly more serious than first-time errors.

What Attorneys Are Actually Doing Wrong

Across the catalogued cases, a few patterns emerge:

1. Using general-purpose LLMs (ChatGPT, Claude, Gemini) for legal research and then trusting the citations they produce without independent verification on Westlaw, Lexis, or another authoritative source.
2. Asking AI to “find supporting cases” for an argument already drafted — which incentivizes the model to invent supportive-sounding case names that don’t exist.
3. Failing to read the cited cases. In multiple cases, attorneys were sanctioned not just for the fake citations but for the fact that even the cases that were real didn’t say what the brief claimed.
4. Junior associates using AI without supervision, partners not reviewing, then partners signing the brief.
5. Repeat conduct after a warning. Courts have shown limited patience for attorneys who get caught once and then do the same thing again.

State Bar Guidance Has Hardened

By mid-2026, over 35 state bar associations have issued formal guidance on AI use. The common elements across these advisories:

• Duty of competence includes understanding the limitations of AI tools you use
• Duty of verification — attorneys must independently confirm every fact and citation in AI-generated work before filing or relying on it
• Duty of confidentiality — attorneys must understand how their AI vendor handles client data, including training-data use
• Duty of supervision — partners are responsible for AI work product produced by associates and staff
• Disclosure obligations to clients in many jurisdictions, particularly when AI is used in billable work

Many federal courts now require affirmative disclosure of AI use in filings, often via standing orders. Check your local rules before filing any AI-assisted brief.

What Tools Actually Reduce This Risk?

The hallucination problem isn’t a reason to avoid AI entirely — it’s a reason to choose AI tools that ground their answers in verified legal content. As of 2026:

• CoCounsel (Thomson Reuters) — every answer cites the underlying Westlaw or Practical Law source. CoCounsel Legal’s Deep Research agents work autonomously across Westlaw’s database.
• Lexis+ with Protégé (formerly Lexis+ AI) — grounded in LexisNexis’s content library with Shepard’s citation validation verifying citations for status and treatment.
• Harvey — used by 100,000+ lawyers across 1,300+ organizations; citation grounding depends on workflow.
• Avoid for citations: ChatGPT, Claude, Gemini — these are useful for drafting, brainstorming, and document analysis, but their citation output is genuinely unreliable and must be independently verified.

Learn more: CoCounsel Review | Lexis+ AI Review | Harvey AI Review

Practical Steps to Avoid Sanctions

If you’re using AI in your practice in 2026:

1. Never trust a citation from a general-purpose LLM. Verify every case on Westlaw or Lexis before relying on it. If it doesn’t exist there, it doesn’t exist.
2. Read the cases you cite. Even when AI-suggested cases are real, they often don’t say what the AI claims they say.
3. Use grounded legal AI tools (CoCounsel, Lexis+ with Protégé) when the work involves citing authority.
4. Disclose AI use where required by your jurisdiction or the court’s standing orders.
5. Document your verification process. When sanctions hearings happen, the record of what you did to check the AI’s work matters.
6. Train junior associates on these rules and supervise their AI use.
7. Re-read your state bar’s AI guidance at least annually — these rules are still evolving fast.

Related Resources

• Risks of Using ChatGPT for Legal Advice
• ChatGPT for Lawyers: Complete Guide
• AI Lawyer in Court: Legal Cases and Implications
• Best AI Lawyer Tools 2026 — citation-grounded tools that reduce this risk

Disclaimer: This article summarizes publicly reported sanctions and bar guidance as of May 2026. It is not legal advice. Consult your jurisdiction’s specific rules and your firm’s risk and ethics counsel before adopting any AI tool in your practice.