Ellis George LLP & K&L Gates: $31,100 Sanctions for Three-AI-Tool Citation Failure | Advocate Prakhar

Case at a Glance

Court	US District Court, Central District of California
Year	2025
Category	AI Hallucination / Multi-Tool / Institutional Sanctions
Jurisdiction	United States (California)
AI Tools Used	Three AI legal research tools (details in court order)
Firms Sanctioned	Ellis George LLP; K&L Gates LLP
Total Sanctions	$31,100

Background

The Ellis George LLP and K&L Gates sanctions represent a qualitative escalation in AI accountability jurisprudence: for the first time, a court confronted a situation where attorneys had used not one but three distinct AI tools in preparing legal research, and the hallucinated citations appeared to arise across the outputs of these different platforms. The Central District of California — one of the busiest federal district courts in the United States — issued an order imposing $31,100 in combined sanctions, making it one of the highest-value AI hallucination sanction orders of 2025.

K&L Gates is a global law firm with over 45 offices; Ellis George LLP is a prominent California litigation boutique. The involvement of two well-resourced firms with sophisticated litigation practices sent a clear signal that no firm is insulated from AI hallucination liability by virtue of its size or reputation.

The AI Issue

The central legal questions were whether the use of multiple AI tools constitutes due diligence or compounds the risk of hallucination; whether the absence of firm-level AI verification policies is an aggravating factor in sanctions determinations; and how sanctions should be apportioned between co-counsel firms that both contributed to the same filing containing fictitious citations.

What the Court Decided

• The court imposed $31,100 in total sanctions across both firms for filing briefs containing hallucinated AI citations [FRCP Rule 11 / court’s inherent authority].
• Using multiple AI tools does not substitute for verification — each tool’s output requires the same independent confirmation against authoritative legal databases.
• The court found that the absence of adequate firm-level AI supervision and verification protocols was a systemic institutional failure, not merely individual attorney error.
• Both firms were required to produce evidence of new AI compliance policies to the court within 30 days of the sanctions order.
• Sanctions were apportioned between the firms based on their relative contribution to the impugned filings.

“The fact that counsel employed multiple AI research platforms does not demonstrate diligence; it demonstrates the absence of any reliable verification process. This court will not accept artificial intelligence output as a substitute for legal research conducted through verified sources.”

— US District Court, C.D. California, 2025

The India Angle

For Indian law firms — particularly the growing number of mid-size and large firms now deploying AI research tools across their practice groups — the Ellis George / K&L Gates sanctions highlight a governance risk that extends beyond individual attorney error to institutional liability.

Indian Law Equivalent

Indian law does not yet have a direct equivalent of FRCP Rule 11’s certification requirement, but the general duty of advocates under the Code of Civil Procedure and the Advocates Act achieves a similar outcome. Order XI of the CPC governs production of documents and discovery; more relevantly, Section 35A of the CPC provides for compensatory costs in respect of false or vexatious claims and defences. A court that discovers fabricated citations in a pleading or written submission could use Section 35A to impose costs — effectively a sanction — on the advocate responsible. The Supreme Court’s inherent powers under Section 151 CPC and Article 142 of the Constitution provide further grounds for courts to act against systemic professional misconduct.

Bar Council Rules

For Indian law firms, the Ellis George case raises a structural question about supervision obligations. The BCI Rules do not directly address AI use, but Rule 33 requires advocates to uphold the dignity of the legal profession in all their dealings. More relevantly, senior advocates in firms have supervisory responsibilities over junior colleagues — in India, as in the US, firm management may bear responsibility for systemic failures to implement adequate AI verification procedures across the practice.

Practical Advice for Indian Advocates

• Law firm managing partners and practice group heads should treat AI governance as a firm-level risk management issue — establish a written AI use policy that mandates verification before any AI-generated citation enters a court document.
• Do not assume that using a “legal-specific” AI tool (as opposed to a general-purpose chatbot) eliminates hallucination risk — dedicated legal AI platforms also hallucinate, particularly for older, less-indexed Indian case law.
• For larger firms, consider designating an AI compliance point of contact responsible for maintaining and updating the firm’s AI verification protocols as the technology evolves.

Quick Takeaways

• $31,100 sanctions — highest multi-tool AI sanction in a single US proceeding as of 2025.
• Three AI tools used, zero verified — volume of tools does not equal verification.
• Firms must now produce AI compliance policies as a condition of the sanctions order.

---

Deep Dive: Institutional Liability and the Firm-Level AI Governance Gap

From Individual Error to Systemic Failure

Most early AI hallucination sanction cases focused on individual attorney conduct — a specific lawyer who used a specific AI tool and submitted a specific set of fictitious citations. The Ellis George / K&L Gates case represents the next stage of this jurisprudence: courts are now looking at the institutional environment within which the AI misconduct occurred, and asking whether the firm’s management and supervision structures were adequate.

This shift has profound implications for law firm governance. A firm can no longer address AI hallucination risk solely through training individual attorneys; it must implement and enforce firm-wide policies that create accountability structures, verification requirements, and audit trails for AI-assisted legal research.

The Multi-Tool Problem

One might assume that using three different AI tools for legal research would reduce hallucination risk — if tool A invents a citation, tools B and C would fail to confirm it, alerting the researcher to the problem. In practice, this assumption is flawed for two reasons. First, AI tools trained on similar datasets may produce similar hallucinations — they share common failure modes. Second, the absence of a systematic approach to cross-referencing AI outputs against verified legal databases means that multiple tools can compound rather than check each other’s errors.

The court’s finding that using multiple AI tools “demonstrates the absence of any reliable verification process” rather than due diligence captures this dynamic precisely. The standard for due diligence in legal citation is verification against official, authoritative sources — not cross-referencing AI tool against AI tool.

The $31,100 Calibration

The quantum of sanctions in AI hallucination cases has followed an escalating trajectory since Mata v. Avianca ($5,000 per attorney, 2023). The Ellis George / K&L Gates order of $31,100 is notable both for its absolute amount and for its apportionment between two firms — suggesting courts are becoming more sophisticated in attributing responsibility and calibrating consequences to the institutional dimension of the misconduct.

Sanctions of this magnitude are significant even for major law firms — they represent real financial consequences that justify the implementation of robust AI compliance frameworks, not merely nominal deterrents that large firms can absorb without behavioural change.

Implications for Legal Malpractice Insurance

A development that has received insufficient attention in AI hallucination coverage is the implications for legal malpractice insurance. If AI citation errors cause adverse outcomes for clients (missed deadlines, lost motions, failed appeals), those clients may have malpractice claims against their lawyers. Insurers are beginning to ask about AI use policies in renewal and underwriting processes. For Indian firms using AI tools, it is worth reviewing whether current professional indemnity insurance covers AI-related claims — most Indian professional indemnity policies predate the generative AI era and may not clearly address this category of risk.