Clio's AI agents can now answer calls and write time entries—but who keeps the record?

The shift from tool to agent

Clio, one of the largest legal practice management platforms, announced on 22 October 2026 that it is rolling out a unified platform with agentic workflows, an AI phone system and a new document management system ^[1]. The AI agents do more than suggest text or summarise cases. They answer calls, draft documents, update matters and log time, all without a lawyer watching every step.

For firms that adopt it, the efficiency gain is real. The question is what happens when something goes wrong. Say a client claims the AI promised a settlement figure the firm never approved, or booked a meeting the lawyer knew nothing about. The client has a recording or a screenshot. What does the firm have?

A receipt for every action

When a person sends an email or answers a call, they remember doing it, and there's usually a sent folder or a call log to check. When an AI agent does it, that memory doesn't exist. The agent acts, the client receives the message, and the firm finds out later, if at all.

It's a bit like handing your credit card to someone else and asking them to keep the receipts. You'll see the statement at the end of the month, but by then the detail is gone. An AI agent that acts on a firm's behalf needs something closer to a point-of-sale receipt: a record of what it did, when it did it and on whose instruction, created at the moment it happened.

That record matters most when it's questioned. A call log shows a call took place. It doesn't show what was said. A draft saved in a document system shows a version existed, but not whether that was the version the client saw, or whether anyone changed it before the dispute started.

Article 50 and the disclosure duty

From 2 August 2026, Article 50 of the AI Act requires deployers of AI systems that interact with people to inform those people that they are interacting with an AI, unless it is obvious from the circumstances ^[2]. The rule applies across the EU, and it covers chatbots, voice systems and any AI that communicates directly with clients or the public.

For law firms using Clio's AI phone system, that means telling callers they're speaking to an AI, and being able to show later that the notice was given. The Act doesn't say how to prove it, but a supervisor or a court that's asked to decide whether the duty was met will want evidence, not assurances.

A transcript isn't enough if the firm controls it. The question isn't just what the AI said, but whether the record the firm produces is the one that was created at the time, unedited.

What this means for you

If your firm is piloting or adopting AI agents, here are three questions worth asking now:

• When an AI agent answers a call, sends a message or updates a client matter, does the system create a checkable record of what it did, or just an internal log the firm could edit?
• If a client disputes what the AI said or did, whose evidence would a court see first—the client's screenshot or recording, or the firm's own timestamped record?
• Does your retainer or terms of service tell clients that AI may handle parts of their matter, and if Article 50 applies, how are you documenting that clients were informed?

The answers don't have to be perfect, but they should exist before the first dispute, not after.

Sealing what the agent did

This is a case where a digital witness fits. The thing to seal is each action the agent takes: the call summary, the instruction it received, the draft it sent or the time entry it logged. Sealing doesn't mean copying the data somewhere else; it means taking a fingerprint of the record and sealing that fingerprint on a chain where any later change would show, even one made by the firm itself.

Anyone the firm shares the record with—an ombudsman, a professional body or a court—can check that it matches the sealed fingerprint. If it does, the record is unchanged. If it doesn't, someone has edited it.

The fingerprint is created when the action happens: when the call ends, when the draft is sent, when the time entry is written. That gives the firm its own receipt, separate from the client's screenshot and from any internal logs the firm could quietly revise.

What it doesn't do is prove the action was right, or that the AI followed instructions, or that the client consented. It shows the record existed at that time and hasn't changed since. That's narrower than it sounds, until the day the firm needs to show it.

The technical side is a short integration. Clio's platform could send a fingerprint of each agent action to WitniumChain through the REST API, the SDK or the MCP server, and get a witness ID back in milliseconds. The content stays in Clio's systems. The fingerprint goes on the chain. To verify, the firm (or anyone it shares the record with) recalculates the fingerprint and checks it against the sealed one. The first 2,000 witnesses a month are free. Read the docs for the integration options.

How it would work

1. The AI agent completes an action (call, draft or time entry).
2. Clio sends a fingerprint of the record to WitniumChain and receives a witness ID.
3. When questioned, the firm shares the record and the witness ID so the other party can verify it.

What to watch

AI agents are moving faster than the guidance on how to supervise them. The AI Act's high-risk requirements, including logging under Article 12, apply to Annex III systems from 2 December 2027 ^[3]. Whether legal AI for law firms counts as high-risk is still debated; most readings say it doesn't, but the question keeps coming up.

Even if it doesn't, the professional duties haven't changed. A lawyer is responsible for the work done in the client's name, whether a person or an AI agent did it. When something is questioned, the burden is on the lawyer to show what happened. That burden is lighter when there's a record nobody had to trust.

Clio's launch is a clear signal that agentic AI is here, and law firms that adopt it early will learn the most. The evidence question—who keeps the record, and can it be checked—should be part of that learning, not an afterthought.