Legal AI: Unbridled benefits or Trojan horse?

By Alex Heshmaty on September 2, 2026

The US administration recently issued a directive with significant, albeit indirect, implications for the adoption of AI by law firms on this side of the Atlantic. The American government ordered Anthropic to revoke access of their latest model (at the time) Fable 5 to all foreign nationals on the basis of national security. Although the ban had little, if any, impact on the UK legal sector, it was a clear demonstration of the danger of becoming reliant on the technology; AI is a resource which can effectively be shut off at the moment’s notice by state actors, irrespective of any contractual obligations. One US legal technology challenged the directive, perhaps out of concern that confidence in the reliability of their services could be undermined by the move. But although this episode illustrates one stark danger of over reliance on AI by legal professionals, there are plenty of other risks.

Reputational damage and data protection

Law firms ending up with egg on their face as a result of submitting court documents replete with “digital hallucinations” is nothing new. However, the practice only seems to be gathering pace, with almost 2,000 known examples and counting, and some leading firms being caught up in the ensuing scandals. Reputational damage aside, courts have also issued fines both in the US and Europe for hallucinations. There is a heavy verification burden to mitigate errors creeping in as a result of using AI tools, and some commentators argue that this can wipe out any efficiency gains, actually causing lawyers to spend more time checking and correcting AI outputs compared to just doing the job themselves without AI. The SRA released a warning notice on the misuse of AI, noting this scourge of fake citations and false information, as well as client confidentiality risks posed by AI platforms. On the latter point, despite any data protection assurances offered by legal AI providers, the fact that the aforementioned US government directive was made purportedly for national security presumably means that they could similarly order the release of private and confidential data submitted to AI tools on similar grounds.

Escalating costs

Initially law firms which adopted AI tools encouraged fee earners to use them as much as possible, with bonus schemes even being tied to volumes of tokens burned through as a result of prompts (a practice extending beyond the legal sector, known as “tokenmaxxing“). But Legora recently announced that it’s moving away from “per seat” billing to consumption based pricing, a fee methodology which is likely to spread amongst all the legal AI products due to the changing economics of AI tokens. Firms may start to notice significant hits to their P&L as a result of rising licensing fees, as well as the costs of training up staff to keep them up to date with the fast changing AI tools.

Existential threat

There are several threats posed by over reliance of law firms on AI, some of which are existential. In a landscape where DIY AI law and AI driven law firms are on the rise, traditional legal practices are at risk of losing not just their clients, but their own talent. As AI is increasingly used for many of the routine tasks originally undertaken by trainee and junior lawyers to hone their skills, some firms may decide to stop hiring junior lawyers, with the rising popularity of consultant model firms being a case in point. But pulling up the drawbridge in traditional firms could lead to law firms becoming staffed solely by senior lawyers and, with nobody left to replace them as they retire, resulting in the business winding up or being acquired and asset stripped.

A separate problem, which potentially impacts the ability of firms to survive if the AI rug is pulled out from under their feet, is the concept of cognitive offloading. The more that great legal minds rely on AI to carry out complex legal analysis, the more that their brains could atrophy, leading to the scenario of once brilliant lawyers becoming more akin to reverse-centaurs, simply rubber stamping the decisions of AI. Many of the complex and technical legal skills could end up being carried out by machines, with human lawyers only being needed to physically attend court, sustain client relationships and win new business.

Integrity of the legal system

Some of the broader challenges facing the legal system as a result of over reliance on AI include a potential threat to judicial independence. Dame Victoria Sharp, president of the King’s Bench Division, recently warned that increased reliance on AI tools by judges could lead to a “gradual drift” towards judicial standardisation based on potentially biased AI outputs, and – in the case of judges who treat AI as sources of judicial reasoning – effectively “outsourcing part of the judicial function” to AI companies.

There is also a threat to the criminal justice system due to the use of AI by the Crown Prosecution Service (CPS). A recent case involved the CPS submitting documents containing AI hallucinated fake citations to the High Court in an extradition appeal. Although this was picked up before having an impact on the hearing, this clearly demonstrates how AI could have a deleterious effect on natural justice, with the judge admitting that it “would be naive to assume that there will not be an increasing use of artificial intelligence in legal work in future”.

Risks of avoiding AI

Despite the many risks of lawyers becoming overly reliant on AI, these need to be weighed up against the consequences of avoiding the technology altogether. One of the obvious downsides is that law firms which harness AI tools may be able to complete work for clients more quickly and at a reduced cost, so firms which opt out of AI are liable to fall behind their competition. Furthermore, clients are increasingly coming to expect (and sometimes demand) reductions in legal fees, as they assume their lawyers are using AI to complete work more efficiently, a situation further complicated by the growing popularity of fixed fees over billable hours. Then there is the problem of “shadow AI“ where lawyers in a firm which has an official no-AI use policy might decide to use AI on their personal devices for work purposes, which makes oversight much more difficult.

A novel risk of choosing to go AI-free is the recent suggestion by the UK Jurisdiction Taskforce (UKJT) that lawyers could be sued for negligence if their avoidance of AI tools results in an unfavourable outcome for a client. In section 7 of its Legal Statement on Liability for AI Harms, it states that a legal professional “could be liable for failing to use AI in circumstances where a competent member of their profession would have done so.” However in the same section, it also states that a lawyer “may be found negligent for using AI inappropriately, for using an unsuitable model, for failing to conduct proper due diligence, or for failing to test AI or validate its outputs effectively.” This statement, which appears to put lawyers between a rock and hard place, has been described by Brian Inkster as “inherently flawed”.

Conclusion

There is clearly a balancing act to be drawn between lawyers harnessing the benefits of AI, whilst at the same time not becoming so reliant on the technology that its sudden absence could dramatically impact their service to clients. There are multiple obstacles which firms need to navigate in choosing how to adapt to a legal marketplace saturated with fast changing AI tools and a whole host of related disruptive forces. Deciding how to engage with this transformative technology in the short term could impact the long term survival of individual firms as well as the legal profession as a whole.