Say You’re a Robot

If your contact center serves anyone in Europe, your chatbot now has a legal obligation to admit what it is.

That took effect this week.

On August 2, the European Commission began enforcing Article 50 of the AI Act. Chatbots, voice assistants, and AI agents have to clearly tell people they’re dealing with a machine rather than a person.

The scope is wider than most US teams assume. The rule reaches any company whose AI output is used in the EU, not only companies based there, and penalties run to €15 million or 3% of worldwide annual turnover, whichever is higher.

The fine is the least interesting part of this.

The disclosure has to actually work.                                       

Under the Commission’s guidelines, burying it in your terms and conditions doesn’t satisfy the requirement. Neither does a metadata tag nobody sees, or calling your bot an “assistant” and hoping the customer infers the rest. The information has to be perceivable inside the interaction itself.

There’s a narrow exception when the interaction is obviously automated, and the test for “obviously” is where this gets interesting.

The standard is a reasonably observant person drawn from your actual audience. And the bar goes higher when children, elderly customers, or people with disabilities are part of that audience.

Read that again as a CX leader rather than a compliance one.

A regulator just wrote a design standard. Your disclosure has to be clear to the people you actually serve, and clarity gets measured by the least advantaged among them.

That’s a tougher test than most of us apply to our own IVR prompts.

We made a version of this argument in Lower-Value Human Capital. The words a company chooses shape whether people trust the thing standing behind them.

Air Canada’s website chatbot gave a customer wrong information about bereavement fares. The airline argued the chatbot was responsible for its own answers. A British Columbia tribunal disagreed. Michael Fisher covered the case in Ardeo Ergo Sum, and it shows why disclosure sits upstream of a much bigger problem.

That case was about accuracy and liability. Disclosure comes earlier in the chain. It’s the moment a customer decides how much weight to put on whatever they’re about to hear.

Which is why the compliance framing undersells this.

Most teams will respond by adding a line of text and moving on. The more useful response is to treat the disclosure as the first design decision in the conversation, because it sets the expectation that everything after it depends on.

Three things a bot should say up front: what it is, what it can handle, and how to reach a human.

A bot that says all three has set honest terms. One that mumbles a disclaimer and then overpromises has built the exact gap customers punish you for.

Kerry Robinson has a related line running through A1B: Customer Zero to AI-First. You own the output. What Europe added this week is that you also have to say where the output came from.

There’s no federal equivalent in the US yet, but a few states already reach customer-service bots. Utah has required disclosure in commercial interactions since 2024, when a customer asks for it, and automatically in health, financial, or legal contexts. Maine went broader in September 2025: tell people they’re not talking to a human whenever a reasonable customer couldn’t tell the difference.

California, New York, Washington, and Connecticut have passed rules too, though those mostly target companion bots rather than the contact center. Healthcare and mental-health bots face stricter rules stilI.Illinois and Nevada bar them from posing as clinicians, while Utah and California require explicit disclosure.

The map matters less than the direction it points. Every one of these laws is asking the question Europe just asked, and your customers were asking it first.

Would your bot pass the test with your oldest customer?

— Damian

Here’s what went down this week.

Bleeding Edge

Early signals you should keep on your radar.

State chatbot laws are moving from transparency to risk management, with Washington’s HB 2225 signed in March 2026 and effective January 1, 2027. The newer statutes are capability-based rather than disclosure-based: if a system can sustain an ongoing relationship with a user, it carries duties whether or not the customer ever asks what it is. Disclosure was the cheap part, so the teams treating this week’s EU deadline as a one-time text change are the ones who will be re-architecting consent and safety flows next year.

Puzzel’s State of Contact Centres 2026 report found that only 3% of contact centers run on a single unified platform, and the average organization now operates 3.9 separate contact center technologies. Every AI agent you deploy has to reach across that estate to be useful, so integration work decides whether a deployment survives production more than model choice does. If a roadmap assumes one platform, it is describing someone else’s contact center.

Leading Edge

Platform changes live now that need a decision this week.    

Genesys Cloud ends native Enhanced TTS support for selected Google and Microsoft voices . Those voices become third-party integrations from here, so keeping them means moving to the Google or Azure TTS integrations in AppFoundry and rebilling under the BYOT-A model. If nobody has checked which flows still call those voices, that audit is this week’s job rather than next quarter’s.

Google Cloud released version 5.0 of its CCAI Platform, adding disposition timing and API-based chat routing. Disposition timing lets an agent attribute wrap-up time, disposition codes, and notes back to a previous interaction rather than the one currently open. It sounds small until you remember that every workforce and quality metric you report upward is built on disposition data landing against the right call.

Future You’s Problem

Handle it now, or explain it later.