Practice #P020 Tech & AI

The model hallucinated. Your liability did not.

"729+ documented AI hallucination incidents in enterprise and legal settings by 2025. Sanctions reaching six figures. California AB 316, in force since January 1, 2026, eliminates the AI-did-it defense. The model's error is your liability. The question is whether you have built the controls that make it defensible."

729+

AI hallucination incidents documented in enterprise and legal settings by close of 2025. Up from 280 in 2024.
(Law360 AI Tracker)

$100,000+

Sanctions for AI hallucination in individual cases by late 2025. From $500 in 2023.
(Artificial Lawyer 2026)

#AI #AI-Liability #AI-Governance #Responsibility

THE DECISION STAKES

The AI did not sign it. You did.

"It is no answer to say that the citation came from an AI tool. Counsel bears personal responsibility for every authority placed before this court." — Justice Saini, UK High Court, Ayinde v Haringey & Al-Haroun v Qatar National Bank, April 2025

AI hallucination liability is not a technology risk. It is a process risk. The four controls below are not technical. They are operational.

Every AI output that reaches a client, a regulator, a court, or a board carries your signature. The verification posture you build now determines whether an error becomes a correction or a sanction.

THE DECISION TOOL

Four moves. One decision you can defend.

VERIFY

Implement mandatory human verification for any AI output that reaches an external party or informs a material decision. Define verification as a process step, not a recommendation. Log who verified, what they checked, and when.

Courts and regulators do not accept 'I did not check.' Verification is not a quality step. It is the liability transfer mechanism from the model to the professional.

SCOPE

Define in writing which tasks AI can perform autonomously, which require human review before use, and which are out of scope entirely. High-stakes outputs: legal, regulatory, clinical, financial. Exclude or gate these explicitly.

A scoped AI use policy protects the organization and gives practitioners clear decision rules. An undefined scope means every use of AI is a personal liability judgment call.

DOCUMENT

Build an audit trail of AI-generated content: which tool, which prompt, which output, who reviewed it, what changes were made before use. Retain for a minimum of 4 years to align with California civil rights AI record requirements.

Documentation is the only defense if an AI error surfaces later. Without it, you cannot prove what the model produced and what your team verified. The absence of a trail is itself evidence of negligence.

INSURE

Review professional indemnity and D&O policies for AI-generated error coverage. Most policies written before 2024 do not cover AI hallucination claims explicitly. Negotiate AI-specific clauses at renewal. Disclose AI use to insurers where required.

The enterprise that has verified, scoped, and documented its AI use is insurable at standard rates. The one that has not is either uncovered or overexposed.

Noland v. Land of the Free

California Court of Appeal, 2nd District. September 2025. Published opinion. AI hallucination in appellate brief.

21/23

Citations in the appellate brief that were fabricated by AI: 21 out of 23. The court imposed a $10,000 individual fine, the highest single sanction to date at time of ruling. The opinion was designated for publication, meaning it now serves as binding precedent in California. (California 2d District Court of Appeal, September 2025)

The case itself was straightforward: an employment dispute, summary judgment affirmed on appeal. What the court could not ignore was the brief: 21 of 23 citations were AI hallucinations. The cases did not exist. The quotes were invented. The court did not sanction the AI tool. It sanctioned the attorney who submitted the brief without verification. The California Judicial Council followed with guidelines requiring judges to either ban generative AI or adopt a formal AI use policy by December 15, 2025. The opinion is now cited in every AI governance discussion in professional services, legal, and enterprise risk

The court did not rule that AI is unreliable. It ruled that using AI without verification is incompatible with professional responsibility. That ruling now applies beyond law firms.

Download the full case

PDF · 5 slides · Free access · Downloaded 0 times

Let's discuss this

Key questions

How do you verify an AI-generated legal brief, financial model, or regulatory submission when the verifier lacks the subject matter expertise to detect a plausible-sounding hallucination?

At what point does a documented AI use policy become a defense in a professional liability claim, and what elements make it legally credible?

When your professional indemnity policy was written before 2024, which AI-generated outputs are currently outside your coverage and how do you know?

Pre-decision checklist

VERIFY — completed

SCOPE — completed

DOCUMENT — completed

INSURE — completed

By Fabrice Macarty

This case resonates?

Mandate human verification for every AI output that leaves the building. Define the scope of AI use in writing before the next deployment. Build the audit trail from day one. Then check your indemnity policy covers what you are already doing.

Start the conversation