AI Hallucinations: A Real Risk for Lawyers
Originally published at 3ital.org — the summary below links to the full piece.
Written with Dr. Roberto Rosas and published at 3ITAL, this article examines the risk that has already ended careers: AI output that seems plausible but is fabricated — delivered confidently, with no signal of uncertainty.
The cautionary landmark is Mata v. Avianca (2023), where lawyers cited six nonexistent, ChatGPT-generated decisions and were sanctioned. But the article’s more uncomfortable finding is that purpose-built legal AI tools are not immune: in the research discussed, such tools erred on 17–34% of test queries, and independent studies found hallucinations in roughly one out of six legal research queries.
The professional-responsibility conclusion is direct: the lawyer’s duty of competence and candor extends to material prepared with AI’s help. The article closes with practical mitigation — independently verify every citation, treat AI as supplementary rather than authoritative, adopt internal usage guidelines, and invest in continuing education on what these tools can and cannot do.