Updates

You have read their factum. You know the cases they cited.

But do you know how they tend to argue when the judge pushes back? The patterns they return to when their primary position softens? The authorities they lean on in this particular type of proceeding?

Most of the time, you find out in the room.

That is not a preparation failure. It is just how litigation has always worked. You prepare your case. You adapt when you get there.

Your accountant went to school for years.

They understand tax strategy. They know which deductions hold up under review. They can see the shape of your business and tell you where it is bleeding quietly.

That is not what most small business owners pay them to do.

Most of the bill goes to categorizing transactions. Matching receipts. Asking you, again, for that one missing invoice from March.

You applied. The role was a fit. You heard nothing.

Most applicants never find out why. The resume looked fine to you. It looked fine to everyone who read it. But the system that processed it first did not agree.

Applicant tracking software filters before any human sees the page. A keyword in the wrong place, a section heading the parser does not recognise, a date format it cannot read. Small things. Quiet things. The kind of thing no one tells you about.

The fix is not a rewrite. It is usually five lines.

The new myth going around legal tech circles is that hallucination is a solved problem.

It is not.

The latest generation of legal AI is better at sounding right. That is different from being right.

A citation that does not exist looks identical to one that does. The case name is plausible. The year is plausible. The holding is close enough to the real law that a quick read feels like confirmation. That is the actual problem. Not the obvious errors. The ones that pass the first read.

You cite three cases. They look right. The citations are formatted correctly. The case names sound familiar.

They are not real.

This is not a hypothetical. Lawyers have been sanctioned, publicly, for filing AI-generated citations that pointed to cases that do not exist in any court record. The AI did not flag the error. It presented fabricated authority with the same confidence it presents real authority.

That is the core problem with most AI tools in legal work. They are built to sound correct. They are not built to be correct.

You kept every receipt this summer.

The gas fill-up before the client site visit. The lunch. The parking. The office supplies you grabbed on a Saturday because you needed them Monday.

They are all in a box. Or a bag. Or a folder you stopped updating in July.

September arrives and the box is still there.

You cite three cases. They look right. The citations are formatted correctly. The case names sound real.

They are not real.

This has happened. It is on the record. Lawyers have filed AI-generated briefs containing fabricated authorities, and courts have noticed. Some of those lawyers are still explaining themselves to their law societies.

The cost is not just embarrassment. It is your reputation, your file, and depending on the day, your licence.