Many enterprises have treated the EU AI Act as a problem for another day. But their procrastination could be about to become costly.
On August 2, the EU AI Act’s transparency requirements under Article 50 begin to apply. While the most complex requirements for certain high-risk AI systems have been pushed back, customer-facing AI cannot be filed away under ‘later.’
In a CX context, if a customer is talking to a chatbot, being assessed through emotion-recognition technology, or having their behavior analyzed by AI, organizations must be completely transparent about the process.
Indeed, in an exclusive interview with CX Today, Kristina Holt, Managing Associate at law firm Foot Anstey, explained this issue:
“You need to clearly identify that you’re talking to an AI, and anything that sort of makes people think they’re talking to a human being needs to be avoided.”
This ruling opens up difficult questions around chatbot design, speech analytics, third-party technology, agent assist, escalation routes, and the increasingly blurry line between a helpful automation and an intrusive one.
The EU AI Act Has Not Been Delayed, It Has Split
The recent EU AI Act changes have created a confusing picture for enterprises.
Yes, the deadline for standalone high-risk AI systems listed in Annex III has moved to December 2, 2027. These are systems used in areas such as recruitment, education, creditworthiness, access to essential services, and employment management.
Meanwhile, requirements for high-risk AI embedded in regulated products, including certain medical devices, vehicles, machinery, and toys, will apply from August 2, 2028.
But those postponements do not erase the August 2026 milestone.
According to the European Commission, Article 50 applies from August 2, bringing transparency obligations for providers and deployers of certain AI systems. These include interactive AI systems, generative AI, deepfakes, emotion recognition, and biometric categorization tools.
For direct customer interactions, providers must design AI systems so people are informed that they are interacting with AI, unless that is already obvious.
The notification should arrive at the first interaction and be clear and distinguishable. That puts customer service AI firmly in view, as Holt detailed:
“The thing around that is transparency. It’s really important. So you need to clearly identify that you’re talking to an AI.”
For many contact centers, this is not a new principle. Plenty of chatbots already announce themselves, while virtual assistants are often given names that make their non-human status reasonably clear.
Yet there is a large gap between “reasonably clear” and a consistent, auditable approach across every customer channel.
That gap is where trouble starts.
Why This Matters for Contact Centers
Most enterprise CX teams will be AI deployers rather than AI providers.
They are not training foundation models or building large language models from scratch. They are putting AI into the customer journey through contact-center platforms, chatbot vendors, CRM systems, speech-analytics tools, and agent-assist applications.
That distinction matters legally. But operationally, it should not become a loophole.
A third-party provider may carry obligations for how its system is built, but the enterprise using the tool still owns the relationship with the customer, the service design, and the consequences when something goes wrong.
Holt argues that “you can’t get rid of your responsibility,” and should only point to your third-party provider if “you are confident that they are in fact handling it.”
That means a contact center cannot simply assume that a vendor’s ‘AI Act-ready’ badge settles the issue.
Take a customer chatbot, for example. The enterprise needs to know whether the AI disclosure appears where the customer actually encounters the tool. It needs to know who controls the wording, how the system behaves after an escalation, and what happens if the chatbot gives an inaccurate, discriminatory, or nonsensical response.
As Holt said: “Who do they [the customer] think they’re interacting with, and who has control over that conversation?”
Those are the questions that matter. Not the marketing claims in a vendor slide deck.
The Chatbot is the Easy Part
The obvious use case is a chatbot.
A customer lands on a support page, starts a conversation, and sees an introductory message telling them they are interacting with an AI assistant. That is the sort of transparency design many companies will now be reviewing ahead of the deadline.
But chatbots are only one part of the modern CX AI stack.
Contact centers are also using AI to summarize calls, recommend next-best actions, guide agents during live conversations, predict customer intent, classify contacts, and analyze voice and text for sentiment.
Some of those tools sit in the background. They do not always respond directly to the customer. That is where the legal and ethical lines become more complicated.
For example, a customer may not know that a system is analyzing their language, tone, or behavior to determine whether they are frustrated, vulnerable, likely to churn, or “in the mood to spend money,” as Holt described it.
“The more you’re interfering and the more you’re doing something that’s intrusive, I guess, in that way of interacting, probably the more notice you need to give.”
That does not mean every use of AI behind the scenes automatically triggers the same Article 50 obligation as a customer-facing chatbot. The Commission’s guidance is specific about the systems and circumstances in scope.
However, organizations should not interpret that distinction as permission to be vague.
There may also be GDPR transparency considerations where personal data is being used for profiling, analytics, or decision-making. The more central the AI is to the outcome, and the more sensitive the inference, the harder it becomes to argue that a brief mention buried in a privacy policy is enough.
Holt’s practical and somewhat colloquial test for CX leaders is to ask themselves, “How creepy is it?”
“If you think, well, actually, people think that was quite creepy, so I probably should tell them. It’s probably the answer to that.”
It is a better test than many enterprises may like to admit.
Transparency Needs to Fit the Customer Journey
A disclosure only works if customers can see it, understand it, and act on it.
That means contact centers should avoid treating Article 50 as a checkbox exercise involving one line of legal text at the bottom of a chat window.
A customer should know from the start whether they are interacting with an AI system. If the system is supporting an agent rather than speaking directly to the customer, organizations should consider how they communicate the broader role AI is playing, particularly where the technology is used for more sensitive analysis.
There is also a strong commercial argument for getting this right.
In a July 2026 study from Exclaimer, 46% of UK adults said they had questioned whether a message was genuine, while 43% said they had questioned the authenticity of a communication more broadly.
The survey, conducted by OnePoll among 1,000 UK adults, found that 53% said the platform used affected their trust in a message. Professional email addresses and full contact details were among the signals people use to judge whether a company communication is real.
The point is not that every AI-assisted email needs a label. Article 50’s provisions around AI-generated text are more targeted than that, and the Commission sets out important exceptions, including text that has undergone meaningful human review or editorial control.
Still, the research demonstrates a wider customer problem. People are already struggling to work out what is authentic.
In that environment, unclear AI use is unlikely to build confidence.
Human Oversight Is Good CX, But It Is Not a New Article 50 Rule
This is where enterprises need to separate legal requirements from sensible practice.
Human oversight, logging, monitoring, and vendor governance are all important for responsible AI deployment. But they should not be presented as brand-new standalone Article 50 duties for every chatbot or customer-service use case.
The AI Act’s transparency requirements are one part of a wider framework. Other obligations may apply depending on the system, its role, and whether it falls into a high-risk category.
Even so, CX leaders would be unwise to isolate transparency from the operational safeguards behind it.
If a customer is told they are talking to AI, what happens when they want a human? If the system produces a poor answer, where does the complaint go? Can the enterprise identify the cause and correct the issue?
An unexplained bad AI interaction can look like a company that does not understand its own policies, does not care about a customer’s problem, or has simply lost control of its service operation.
By contrast, clear disclosure can help customers identify what went wrong and give the business a chance to put it right.
As Holt explained: “The customer can come up to you and say, ‘Actually, excuse me, your thing has just done this. It doesn’t make any sense, and you need to sort out what’s happened.’”
For organizations deploying AI agents with the ability to take action across systems, the stakes rise further.
A chatbot that retrieves a knowledge-base answer is very different from an AI agent that can change account details, process a refund, alter an order, or access enterprise systems, as Holt explained:
“Have you got an agent that’s doing stuff? Yeah, because if you have, they’re at much higher risk than if you’re just you know get information.”
That should shape governance decisions, even where the use case is not yet classified as high risk under the Act.
Five Questions CX Leaders Should Ask Before August 2
There is still time for contact center and CX leaders to act. But they need to focus on their live customer journeys, not just the policy documents sitting with legal teams.
- First, identify every AI system that directly interacts with customers. This includes chatbots, voicebots, virtual assistants, and automated messaging tools.
- Second, review whether customers are clearly told when they are interacting with AI. The notice should appear at the beginning of the interaction, not after the customer has already shared information or made a decision based on the conversation.
- Third, map AI systems that affect customers without directly speaking to them. This includes sentiment analysis, biometric categorization, emotion-recognition tools, AI-driven profiling, and automated decision support.
- Fourth, challenge third-party suppliers. Ask what obligations they are meeting, what instructions they provide to deployers, what the contract says about responsibilities, and how they will support incident escalation.
- Finally, check the humans around the technology. AI literacy obligations have applied since February 2025, so staff using AI should understand what the tools can and cannot do, how to question an output, and when to escalate a concern.
The EU AI Act may be moving in stages, but customer-facing transparency is no longer a future problem.
Holt captured the practical opportunity, stating: “It’s probably better just be upfront about what you’re up to.”
For contact centers, that may be the simplest way to approach the deadline.
Make it clear when AI is in the conversation. Make sure customers can reach a person when the situation calls for one. And do not assume a technology supplier has taken care of the hard part simply because it built the tool.
The deadline is here. The customer will notice if the organization is not ready.
