Why would AI mistakes cost lawyers more than other professionals who also use AI tools?

This explores why provenance tracking (knowing which source each AI-generated claim came from) seems to matter so much in legal work, and whether lawyers really need it more than others or just feel its absence sooner. The corpus doesn't support the idea that lawyers are uniquely special. What it does show is that legal work sits where three pressures meet: the person is personally accountable, the work product is largely a chain of references, and errors get caught in public by a judge or the other side.

The clearest evidence comes from interviews with lawyers using GenAI summaries Does GenAI actually save lawyers time on fact verification?. The summaries looked efficient. But because lawyers stay responsible for every fact they rely on, they had to trace each unclear claim back to its source, and that took longer than doing the work by hand. The key point is that the problem was opacity, not just error rates. A tool that is usually right but can't show its work still forces a full re-check. This is why a legal brief differs from a marketing draft: in a brief, the reference is the substance. A claim with no source behind it has no value, however fluent it reads.

The consequences show up in court records. In a review of 114 US cases with suspected AI errors, 90% involved solo or small firms Do small law firms misuse AI more often than large ones?. That figure covers errors that were caught, not how often each kind of firm misuses AI, but it shows the weak point: practices without the staff to re-verify are the ones that end up in front of a judge. Something less obvious makes the problem worse. Weaker models tend to drop content, which you can see, while frontier models tend to change it subtly while the document still looks intact Does model capability change how documents degrade?. So better models can make provenance more necessary, not less, because their mistakes are harder to spot by eye.

The comparison with other fields is revealing. Newsrooms reached the same conclusion: in the Data2Story work, binding every number and quote to its origin was what made agent-written stories acceptable to editors, more than polished writing Can source traceability make AI writing trustworthy?. Medicine looks different. Clinicians using an LLM built more complete diagnostic lists than clinicians using search alone Does LLM assistance help clinicians build better differentials?. There the AI widens the set of possibilities, and the doctor's own judgment and the patient's actual condition act as the check. A lawyer has no equivalent check outside the sources themselves. The text is the evidence.

If you want to see what good provenance looks like in practice, evidence selection that explains its reasoning (the AI says why it chose each passage) beat plain similarity ranking by 33% on legal, financial and academic documents while using half as much retrieved text Can rationale-driven selection beat similarity re-ranking for evidence?. So the honest answer is that lawyers don't need provenance more than journalists or auditors do. They're simply the profession where missing it costs the most and shows up the fastest. That makes legal work an early test case for every field where a person has to sign their name to what the AI produced.