������Ƶ

The law can be polite, but it’s rarely soft. Ask a lawyer what they value most and they are likely to say accuracy, evidence, and (probably) coffee. “Politeness”, as important as it is, doesn’t usually make the cut. Precision matters more than pleasantries.

I was reading recently about how AI models are designed. They are generally built with a clear hierarchy for their responses: be harmless first, helpful second and accurate third. Lovely for bedtime stories. Perfect for writing sonnets in the style of Shakespeare. Vital, one may argue, for writing a speech in the tone of Boris Johnson, Donald Trump or Nigel Farage. 

Less useful, however, when you’re drafting a clause that could trigger a £30 million liability. In law, accuracy isn't optional. It's mandatory

So it’s a curious thing that lawyers are turning to AI tools that are not designed for the legal use case. ChatGPT, for example, is hard coded to be more concerned with being harmless than being helpful. It is mindful at all times that a strongly worded indemnity clause might hurt someone’s feelings.

Being harmless can be harming

It’s not that harmlessness is bad. Nobody wants an AI that rants about legal conspiracies or rewrites an NDA in iambic pentameter (mental note, might give that a go). No one’s advocating for AI that blurts out legal advice like a trainee on two hours’ sleep and a misguided sense of confidence. But when the need to avoid controversy overrides the need to convey nuance, the quality of the legal advice and direction is at risk. Half-truths dressed up as risk mitigation can do more damage than directness.

Ask a non-legally trained AI model to draft something important, like a clause that shifts material risk. It might hand you back something so vague you could drive an uninsured truck through it. Elegant, yes. Accurate? I’m not so sure. Over-cautious AI can edit the edges off your advice. It hedges. It sanitises. It rewrites “material breach” as “a bit of a hiccup”. That’s when things get dangerous. Vagueness doesn’t lower liability. It raises it.

Lexis+ AI doesn’t play that game. It delivers answers sourced from the full weight of ������Ƶ content including primary law, secondary commentary, Practical Guidance and expert analysis - all linked, all verifiable and all searchable from a single prompt. Accuracy and truth are vital to the law. 

Critically, the legal profession already possesses an effective safeguard - human oversight. Lawyers are highly trained to manage sensitive, complex information responsibly. Unlike consumer-facing AI tools, legal AI systems need not prioritise harmlessness to the same degree, precisely because their outputs are mediated by expert judgment. This context significantly reduces the risk of harm.

Lexis+ AI is built for these lawyers. Every response is grounded in the full weight of ������Ƶ’ legal and practical content.  You don’t just get the law, you get the context, the commentary and the confidence to act on it

Truthfulness, with footnotes

Now, let’s deal with the obvious question. If an AI answer is blunt or impolite, does that create more risk?

Not really. Not if it’s built like Lexis+ AI - where every output comes with direct citations to the underlying source, and every session is protected by enterprise-grade security. It doesn’t just give you an answer - it shows its working. If something doesn’t feel quite right, you can trace it back to the statute, case, or guidance that underpins it. 

Factual correctness is the strongest shield against malpractice claims. When AI-generated content is consistently accurate, defensible and reliable, it significantly reduces professional risk.

The real danger, in my view, arises when critical details are obscured in the pursuit of harmlessness, leading inadvertently to incomplete or inaccurate legal work.

This is what makes Lexis+ AI so powerful and defensible. It’s grounded on real law, fine-tuned daily by legal experts and constantly improved through expert review. It gets smarter the more it works. It’s about as close as you will ever get to a model that equally balances harmlessness, helpfulness and accuracy.

Of course I'd like to know more about Lexis+ AI