The pieces cite several concrete cases. In a U.S. Tax Court matter, a taxpayer's filing included citations the court found were likely fabricated. The judge rejected the argument and warned of possible sanctions, but none were imposed. Deloitte, in a 2025 report to the Australian government, included nonexistent academic citations and a fabricated quotation attributed to a federal court judgment; the firm later issued a revised version and partly refunded its fees. In a defamation suit over Google AI summaries, TV producer Robby Keene saw a court in September 2026 let some claims proceed without deciding whether Google was liable. Whether any of these outcomes sets a lasting rule on liability is not yet settled.
It is not settled who answers for AI errors, and courts and regulators are still testing the question as AI use becomes routine in professional work. The core lesson in the pieces is verification: AI errors are hard to catch because they can include plausible citations, and the National Institute of Standards and Technology's 2024 term "confabulations" describes confident but false output. Attorneys should treat citations and factual claims from AI tools as unverified until checked, review firm policies on AI use and sign-off, and assume their organization, not the vendor, carries the exposure.