Compliance officers spend a good part of their year reconstructing the past. The SRA wants to understand how the firm has been using AI, and the honest answer usually lives in scattered places — fee-earners' memories, email threads, a version history here, a separate tool's log there, if that log still exists. It is a half-day exercise once a year, done from rough notes, with gaps you can feel but cannot always name.

We wanted to turn that into something you click.

What the check does

Open a matter where the use of AI has started to mount up — drafts produced, summaries pulled, facts extracted and accepted. Click the compliance check. What comes back is not a template with your firm's name dropped in. It is a posture report against the SRA Code of Conduct: which Principles the work on this matter touches, where the firm's published position on AI use is consistent with what actually happened here, and where a COLP would want a note.

Because it reads the matter, it can be specific. It is not telling you what firms in general should watch for. It is telling you what this matter, with its real activity, looks like against the Code.

Where the evidence comes from

This only works because the product already keeps an honest record. Every fact extraction you accepted, every privilege-gate decision, every counterparty pull — those are already logged as the matter moves. The compliance check reads that trail. So the SRA posture report is evidenced from real activity, not reconstructed from memory under pressure.

Once a year, the same module drafts the Annual AI-Use Report your firm needs to evidence — again, built from the actual activity across your matters rather than a form you fill in from what you can recall. You read it, you edit the parts that call for partner judgement, you sign, you file. The machine assembles the evidence; you supply the judgement.

The gate that runs alongside

There is a second thing happening on every matter, quietly, all the time. Every consequential AI action — anything that writes to the matter file, sends a draft externally, or files a court form — pauses at a pre-action gate. You see exactly what the assistant proposes to do, and you approve or reject before it executes.

This is the human-oversight obligation that the EU AI Act, at Article 14, puts on professional users of AI systems. What matters here is that the gate is on by default, for every firm, from the first matter you open. It is not buried inside an enterprise tier or listed as something on the roadmap. The audit log captures each decision, per matter and per fee-earner, so the oversight is not just performed — it is evidenced.

Honest about what is live

The check runs when you click it today. A fully reactive mode — one that re-runs the posture read automatically on every material change to a matter — is the next step, gated on our agent runtime. We would rather ship the click-to-run version that genuinely works than promise a continuous one before it holds.

A generic AI tool cannot do this, because it does not hold the matter. It has no record of what your firm actually did, so the best it can offer is a template and a hope. A compliance officer that reads the matter is only possible because the whole product was built, from the ground up, around the matter — by people who have had to sit across from a regulator and account for their firm.

← All posts