The count keeps rising. An AI hallucination case tracker recorded about 1,598 cases worldwide as at 9 June 2026, and the number has grown steadily as more lawyers reach for these tools without verifying before they file.
What the numbers show
Of those cases, 59 were in the United Kingdom, with the majority in the United States. Most involved fabricated case law, which is the failure that AI produces most readily and that a court spots most quickly. A general model does not know the law. It predicts text that reads like law, and when it has no real authority to draw on, it invents one that looks convincing on the page. The fifty-nine UK cases are the ones that reached a courtroom. The number of drafts that carried a fabricated citation and were caught, or were not, sits well above that.
The trend is the part worth holding onto rather than any single figure. A tracker that is counting in the thousands, and rising, tells you this is not a run of isolated lapses by careless individuals. It is a predictable consequence of a tool being adopted faster than the habits that make it safe. Every firm now reaching for these models is a candidate for the same mistake unless it puts something in place to stop it, and the courts have made clear they will keep naming the firms that do not.
UK court guidance has been direct about the consequence. An unverified AI submission breaches a solicitor's duty to the court. That is not a matter of etiquette or good practice. It goes to the obligation that sits at the centre of the work, the duty not to mislead the court, and it does not soften because a machine produced the words.
Build the habit
The answer is a habit rather than a policy document, though a policy helps. Never file or send AI output that a person has not checked against the source. Treat every citation as unverified until someone has pulled the authority and confirmed that it says what the draft claims. The check is quick when the case is real. It is the fabricated case, the one that cannot be found, that the habit is designed to catch, and that is exactly the one that does the damage if it slips through.
The discipline is simple to describe and easy to skip under pressure, which is why it has to be built into how the firm works rather than left to individual conscience. A tired lawyer at the end of a long day is the person most likely to trust a citation without checking it, and the court will not accept the workload as an excuse.
There is a further reason the habit has to be firm-wide rather than personal. The lawyer who drafted the passage is the one least able to see the fabrication in it, because the confident phrasing of the model has already persuaded them. A second person, or a fixed step that forces the drafter to open the authority and read it, breaks that spell. The check works precisely because it does not rely on the drafter's own sense that the citation looks right, since looking right is exactly what a good hallucination does.
What a firm should do
Put a check step between the AI draft and the filing, and make it non-optional. Name who performs it, decide how it is recorded, and refuse to let anything go out until the step is done. A verification habit costs minutes and protects the one thing a firm cannot afford to lose, which is its standing with the court. As the case count climbs, the firms that stay out of it will be the ones that made the check a rule rather than a hope.
The judiciary’s AI guidance leaves no doubt about how the courts see unverified citations.
If your firm files court documents and the checking rule lives in people's heads, put it on paper with our help before the tracker grows again: get in touch.
