Reported court cases involving AI-generated errors declined in the second and third quarters of 2026, reversing the rapid growth recorded since generative AI became widely available.
The finding comes from legal researcher Damien Charlotin’s database, updated on 23 October. It identifies 2,207 decisions worldwide in which courts or tribunals addressed hallucinated material, including fabricated or misrepresented authorities. Reuters reports that at least 1,500 entries involve US state and federal courts.
The decline should be interpreted cautiously. The database records detected cases that produced judicial decisions—not every erroneous filing. Charlotin says the apparent improvement could reflect better models and greater user caution, but also reduced reporting as AI errors become more familiar. Courts continue to encounter new cases, particularly involving self-represented litigants.
The development matters because it provides an early indication that verification rules, professional awareness and technical improvements may be having an effect. It does not establish which factor is responsible—or that hallucination risk is approaching zero.
For Australian practitioners, the lesson is that improving model reliability does not displace professional verification. Indeed, increasingly accurate systems may create a different risk: lawyers becoming less likely to check exceptional wrong outputs. Verification controls must therefore remain systematic rather than dependent on whether an answer appears convincing.
Sources: Charlotin’s database; Reuters.
Leave a Reply