Our Work · REVIEW
Mata v. Avianca: When an AI Answer Looks Like Legal Research
Does AI-assisted legal research survive direct source verification?
Decision question
Does AI-assisted legal research survive direct source verification?
What the public record supported
In Mata v. Avianca, it did not. The court found that non-existent judicial opinions with fake quotations and citations were submitted and that the lawyers failed their gatekeeping responsibilities after the authorities were challenged. The case is a clear demonstration that fluent output is not authority and that source verification must occur before reliance.
Material findings
- The sanctions order records that non-existent judicial opinions and fake citations were submitted to the court.
- The court expressly stated that there is nothing inherently improper about using a reliable AI tool for assistance; the failure was the abandonment of professional gatekeeping and verification.
- The court imposed sanctions and required remedial notifications to the client and judges falsely named in the fabricated opinions.
- The useful control lesson is not “never use AI”; it is “locate, authenticate and read the underlying authority before relying on the output.”
What Aperture examined
- The sanctions record in Mata v. Avianca and the evidentiary failure created by fabricated authorities.
- The difference between fluent legal text and authentic legal authority.
- Verification controls for AI-assisted legal research.
- Professional reliance boundaries demonstrated by the case.
Boundaries
- Legal advice on any current matter.
- An assessment of professional liability beyond the public sanctions record.
- A general claim that all generative AI legal research is unreliable.
- Jurisdiction-specific ethics advice.
How this demonstrates Aperture capability
This demonstration shows how Aperture frames a decision question, traces material claims to external evidence, separates what is established from what remains uncertain, and keeps the reliance boundary visible. It is designed to demonstrate the research and evidence method rather than imply a client engagement.
Relevant buyer context: Counsel, legal teams and professional advisers.
Scenario: AI-assisted research verification.
Evidence snapshot
12 cited sources · 18 controlled propositions · 39-page PDF snapshot
The PDF is retained as an optional dated snapshot. For current public presentation, use this web page.