# AI, And Experts, And Hallucinations, Oh My!
Lawyers face mounting liability exposure when deploying artificial intelligence tools without understanding their limitations. This reality stems not from novel legal duties but from centuries-old professional obligations that remain fully enforceable in the era of machine learning.
The core tension is straightforward. AI language models generate text with confidence regardless of accuracy. These systems, built on pattern recognition rather than knowledge retrieval, produce plausible-sounding but fabricated information at rates that vary unpredictably. The legal profession calls this problem "hallucination." Courts and bar associations are beginning to treat lawyer ignorance about AI capabilities as professional negligence.
Recent disciplinary actions establish the baseline. In 2023, a federal judge sanctioned lawyers at Levidow Levidow & Oberman for submitting court filings citing fake cases generated by ChatGPT. The attorneys argued they trusted the AI system's output. The court rejected this defense entirely. Judge Hellerstein ordered the lawyers to pay opposing counsel's legal fees and attend ethics seminars. This sanction did not turn on a new rule. It applied Model Rule 8.4 (conduct prejudicial to the administration of justice) and basic competence standards from Rule 1.1 (lawyer shall provide competent representation).
Rule 1.1 requires lawyers to maintain knowledge of legal methods and procedures relevant to their practice. Courts now interpret this rule to include understanding AI tools deployed in legal work. The American Bar Association's formal opinions on AI competence establish that lawyers must comprehend how their tools function, including failure modes. Ignorance of how an AI generates results does not shield a lawyer from malpractice or discipline.
The practical consequence reshapes due diligence obligations. Before using AI for legal work, lawyers must verify outputs against primary sources. They must understand the model's training data, its known limitations, and the tasks for which it remains unreliable. For predictive tasks (assessing case outcomes, analyzing statutes), lawyers need to grasp that AI systems trained on historical data reproduce historical patterns, including bias. For generative tasks (drafting motions, summarizing discovery), lawyers must independently verify every citation and factual assertion.
This burden falls heaviest on solo practitioners and small firms lacking AI expertise. Bar associations across multiple states have published guidance recommending lawyers either develop internal competency or retain consulting experts. Some jurisdictions now treat the failure to disclose AI use to courts as a separate violation. Other jurisdictions remain silent, creating uncertainty about disclosure obligations.
Insurance carriers respond by refusing coverage for AI-generated work or imposing exclusions. Malpractice insurers demand proof of verification protocols. This shifts cost and liability downstream to the lawyer.
The ethical landscape continues hardening. Disciplinary counsel in New York, California, and Florida have opened investigations into lawyers using AI without understanding its operation. State bars debate whether "AI use" itself requires specific client consent disclosures beyond the general competence duty.
For in-house counsel, the exposure intensifies. Corporate clients expect legal departments to leverage AI for efficiency. Yet the in-house lawyer remains individually responsible under the state bar rules where they are licensed. Corporate policies favoring speed and cost reduction do not override professional obligations.
The takeaway reflects established law applied to new tools. Lawyers bear affirmative duties to understand the systems they deploy. These duties predate AI by decades. Technology changes. Professional responsibility standards do not.