The AML gap is in the file.

In anti-money laundering work, we think AI for law firms is most useful when it builds the matter file, not when it judges the client. It should gather the identity check, the source of funds evidence and the risk assessment for each matter, link each one to the document behind it, and show plainly what is missing. The fee-earner still decides whether to act.

The latest figures from the Solicitors Regulation Authority show why the file matters. As Legal Futures reports, the SRA carried out a record 966 AML compliance assessments in the year to 6 April 2026, and the proportion of firms assessed as non-compliant fell from 32% to 22%. Solicitors were still collectively fined £2.6m. Regulatory settlement agreements alone accounted for 145 fines, totalling £2m.

The gaps the regulator found are gaps in the file

The file-level figures are the useful part. Out of 6,592 files reviewed, 5% did not contain evidence that the client had been identified and verified, and 9% did not have any source of funds checks. Files with non-existent or inadequate client and matter risk assessments fell from 16% to 9%.

That is progress. It is also a description of files that do not show the work. We would expect that some of those checks were never done, and that others were done and never written down. Read from the file alone, the two look the same.

The same report points the same way on sanctions. Legal Futures notes that cases relating to a failure to comply with sanctions regimes have tripled, mostly due to human or administrative errors within firms. That is one regulator’s account of one year. It is still a reminder that the failure is often clerical before it is criminal.

Gathering documents is not scrutinising them

The figure we would put on a partner’s desk is this one: where documents had been gathered, the source of funds had not been scrutinised in 18% of files. The documents were on file. The scrutiny was not.

Alexandra Jones, the SRA’s director of AML, said that source of funds enquiries, risk assessment processes and ongoing monitoring “continue to be areas where further improvement is required”. We think a tool that collects documents and ticks a box would make that problem worse. It produces a fuller folder and the same empty conclusion.

What a system worth trusting should do

Every check should point to its document. Each line of the risk assessment, each identity check and each source of funds conclusion should link to the page it relies on. Ask your vendor to open one matter and show you where the evidence for every conclusion sits.

A gap should read as a gap. If there is no evidence for where the money came from, the file should say so in plain words. A field filled with something plausible is worse than an empty one, because a reviewer will pass it.

Scrutiny should stay with a person. A system worth trusting may put the evidence in front of the fee-earner, but it should not write the conclusion for them. The fee-earner writes it, in their own words, and puts their name to it.

Client data should stay the firm’s. In our view an AML file is among the most sensitive things a firm holds. A system worth trusting should never train a model on it, and you should ask for that in writing.

Where to start

Take a handful of closed matters and ask one thing of each: whether a stranger could find the risk assessment, and the evidence behind the source of funds conclusion, without help. Where the answer is no, the gap is a record, and records are the part software can help with.

That is how we think about legal work at Syntexia: every result points back to its source document, and a person decides. If you would like to talk through how such a system is built, tell us.