AI is now part of the legal ecosystem. Lawyers use it to summarize documents, organize timelines, identify gaps, draft questions, test arguments and more.
The question is no longer whether AI can be used in legal work. It already is. The more important question is whether lawyers and their investigative partners are treating AI-generated information with the same discipline they would apply to any other intelligence source.
Recent cases involving attorneys sanctioned or reprimanded for citing AI-generated fake cases show the danger of getting that answer wrong. In several matters, lawyers submitted filings including non-existent legal authorities generated by AI tools. Courts responded with sanctions, fines, professional embarrassment and, in some cases, worse.
Those cases are often framed as cautionary tales about AI hallucinations. That’s true, but incomplete.
The deeper issue isn’t that AI produced bad information. Human sources produce bad information, too. Databases contain errors, and public records can be incomplete, misfiled, outdated or misinterpreted. Social media can suggest connections that are tenuous at best.
The risk arises when AI output is treated as verified fact instead of what it often is: an untested lead.
Intelligence work has always required judgment about source reliability, corroboration and confidence. In legal intelligence, information is rarely useful simply because it exists. It becomes useful when the team can explain where it came from, how reliable it is and whether it can be corroborated. That same standard should apply to AI-generated information.
If an AI tool suggests a person has a criminal conviction, that output should not be treated as a final finding. It should prompt a deeper check against court records, docket databases or other primary materials. The same is true for corporate affiliations, asset searches, property records, social media connections, litigation histories or reputational claims. AI can help identify patterns, surface possible connections and frame follow-up questions. But the closer the information gets to a legal filing, subpoena, allegation, witness strategy or courtroom, the more rigorous the verification process must become.
When a lawyer’s license or a client’s reputation is on the line, “AI said so” is not a defensible methodology.
Early-stage background research may involve a broader universe of leads. At that stage, AI may help identify areas worth investigating, organize large volumes of information or suggest questions the legal team should explore.
Evidence collection and courtroom presentation are different. Anything submitted to a court, used to support a subpoena, presented as a factual allegation or relied upon to establish ownership, intent, damages, credibility or liability must meet a much higher standard.
That’s where tradecraft becomes critical. A disciplined process should ask:
- How will the information be used?
- What happens if it’s wrong?
- Can it be independently corroborated?
- Should it be treated as high-confidence, moderate-confidence or low-confidence?
- Can the team reconstruct how the conclusion was reached?
One of the biggest mistakes is presenting uncertain information as certain. AI-generated text can sound polished and authoritative, but tone confidence is not the same as confidence in source reliability.
AI doesn’t eliminate the need for analysts. It raises the standard. The attorneys who faced sanctions for AI hallucinations were not punished for using new technology. They were punished because they failed to verify what they submitted, failed to recognize the limitations of the source and failed to meet the professional standard required of legal work.
The issue is not whether AI is low-risk or high-risk. The issue is whether the team has the discipline to know the difference between a lead, a finding and a fact.