Law firms have bought into AI. Now lawyers are worried it could be wrong

The conversation about artificial intelligence in law firms has moved on. A year or two ago, the question was whether lawyers would actually use it. Now the question is whether they can believe what it tells them.

That is an important shift. Firms have spent money, signed licences, trained staff and woven AI into workflows. But a new study by LexisNexis, reported in its analysis of confidence in legal AI, found that 85% of legal professionals at large law firms are concerned about relying on inaccurate or fabricated information generated by AI.

In other words, the technology has arrived in the building. Trust has not.

Why accuracy matters more than speed in legal AI

For most industries, an AI tool that is fast but occasionally wrong is still useful. Law does not work like that. An incorrect case citation, a missed contractual clause or a misstated point of evidence does not just slow a lawyer down — it can change the advice a client receives.

This is why AI hallucinations have become such a focus in legal services. Generative AI can produce text that sounds authoritative even when the underlying proposition is false. The real danger is not an obviously absurd answer. It is an answer that is almost right.

A fabricated case citation presented convincingly alongside real authorities can take time, skill and verification to spot. That is exactly the kind of mistake a busy lawyer might miss during a pressured working day.

The courts have already seen what can go wrong

This is not a hypothetical risk. In the joined Divisional Court cases of Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank, the court considered the consequences of false or suspected AI-generated legal authorities being placed before it.

The judgment made the position clear. AI can assist with legal work, but responsibility for that work stays with the professional. A lawyer cannot hand professional judgment to a chatbot and hope for the best.

As LexisNexis noted in its summary of the cases, the episode is a practical warning about verification rather than a reason to abandon AI altogether.

Verification can eat into the productivity gain

There is a practical tension here. The main selling point of legal AI is efficiency. If a system can produce a first draft in five minutes, it potentially saves hours of manual work.

But the output still needs checking. The more time lawyers spend verifying citations, checking factual assertions and comparing AI-generated analysis against primary sources, the smaller the net saving becomes.

That does not make the tool useless. A draft that takes five minutes to generate and 30 minutes to review can still be far more efficient than starting from a blank page. It does mean, though, that AI-assisted legal work is not the same as unsupervised legal work. Firms need to budget for the checking as well as the generation.

Firms need processes, not just licences

The 85% concern rate is a reminder that buying an AI product is not the same as having an AI strategy. Firms need to decide which tasks are appropriate for AI assistance. They need policies on confidential information, guidance on which outputs require verification, and clear rules about who is responsible when things go wrong.

Should a trainee's AI-assisted research always be checked by a supervisor? Does an AI-drafted advice need the same review as a manually drafted one? Can particular tools be used with client documents? These are management questions, not technology questions.

Specialist legal AI could change the confidence problem

Not every AI system carries the same risk. One of the more interesting developments in legal technology has been the rise of specialist platforms built around curated legal information and established professional databases.

The difference between a general-purpose chatbot and a specialist legal AI platform may turn out to be crucial. The question is no longer just which system produces the most impressive answer. It is whether a lawyer can understand where that answer came from and verify it.

Traceability may end up mattering as much as speed.

Caution is not resistance

An 85% concern rate could be read as lawyers dragging their feet. It could just as easily be read as lawyers doing their jobs. Legal services is a regulated profession built on judgment, accuracy and accountability. Caution is not necessarily a barrier to innovation — it may be a precondition for using it responsibly.

The profession has largely moved past asking whether AI belongs in law firms. The harder question is how to use it in a way that lawyers are willing to rely on.

Firms that solve that problem will not just be the ones with the best technology. They will be the ones with the clearest processes for making sure the technology is right.