6 AI Slipups That Got Attorneys Sanctioned in 2025

AI Slipups Got Lawyers Sanctioned - Here's How to Keep It Out of Your Filings

Courts are tired of fake citations and invented quotes. If AI touched your brief, judges expect you to verify every line like you would from a junior associate. Sanctions in recent years have made one thing clear: diligence still wins.

Why courts are sanctioning AI misuse

The rules haven't changed; the tools have. You're responsible for what you file, even if software helped draft it.

• Rule 11 (FRCP): You certify filings are grounded in law and fact after a reasonable inquiry. See the standard: LII: Fed. R. Civ. P. 11.
• ABA Model Rule 1.1 (Competence): Tech competence includes knowing the limits of AI. ABA Model Rule 1.1.
• Rule 3.3 (Candor to the Tribunal): No false statements or bogus citations-period.
• Rule 26(g): Same duty of reasonable inquiry in discovery responses.

Where AI goes wrong in legal work

• Fabricated case citations or docket entries that look plausible but don't exist.
• Misquoted holdings or pin cites pulled from thin air.
• Jurisdiction mix-ups, outdated law, or overconfident summaries.
• Confidential data leakage from past prompts reused in new matters.
• Faulty redactions or auto-summarized facts that omit key nuance.

A six-point playbook to avoid sanctions

• 1) Source-first drafting: Pull law from trusted databases. If AI drafts, it must cite to sources you can independently retrieve.
• 2) Mandatory verification: Every citation gets Shepardized/KeyCited, quotes checked against the original, and docket items confirmed.
• 3) Disclosure protocol: If your court requires it, include a short certification of human review. Don't volunteer more than local rules require.
• 4) Data hygiene: No client or confidential data into public models. Use approved, enterprise tools with retention controls.
• 5) Logs and reproducibility: Keep a record of prompts, versions, and sources relied upon. If challenged, you can show your diligence.
• 6) Train your team: Partners, associates, and staff need the same standards. One weak link can sink a filing.

Verification workflow for briefs (use it every time)

• Collect authorities from primary sources; flag each proposition of law.
• Run citators and note negative history and jurisdiction limits.
• Open the PDF or slip opinion; confirm the exact quote and pin cite.
• Rebuild string cites with only confirmed authorities.
• Scan for "AI tells": hyper-generic phrasing, perfect symmetry, or cites you don't recognize.
• Final pass by a second reviewer who did not draft.

Policies to put in place this quarter

• Approved tools list: Name the allowed models, versions, and settings.
• Use cases: Allowed (brainstorming, outlines, proofreading). Prohibited (final cites, quoting cases without source, confidential facts in public tools).
• Certification language: Short, consistent statement for courts that require it.
• Protective orders and ESI: Ban public AI tools for discovery review and redaction.
• Vendor terms: No training on your data, clear retention, audit rights.

If the court asks whether AI was used

Answer directly and briefly. Emphasize human verification, cite checks, and adherence to Rule 11 and local requirements. Don't overshare process details that aren't requested.

What to do if a hallucination slips through

• Own it fast: file a notice, correct the record, and explain your verification steps.
• Show remediation: updated policy, added checks, and training completed.
• Avoid excuses about "the tool." Responsibility sits with counsel.

Useful references

• Federal Rule of Civil Procedure 11
• ABA Model Rule 1.1 and Comment on Technology Competence

Upskill your team

If your firm is formalizing AI standards, make training part of onboarding and annual refreshers. Useful starting point here: Complete AI Training: Courses by Job.