Start a project
Start a project

Article 50 of the AI Act for insurance chatbots and claims agents

What the AI Act transparency duties require of insurers and MGAs using chatbots and claims agents, and how to build them into a claims flow.

EU AI Act, Insurance, Claims, Transparency

Since 2 August 2026, an AI system that talks directly to policyholders must be designed so they know they are dealing with AI, and the notice must reach them clearly at the first interaction at the latest. For most insurers and MGAs that means a disclosure in every claims chat, voice line and email channel an AI agent touches, plus machine-readable marking of the text it generates, which systems already on the market must have by 2 December 2026.

This guide covers what Article 50 of the AI Act asks of an insurance claims flow, who carries each duty, and how to meet it without making the experience worse for the customer.

What Article 50 covers

Article 50 is the AI Act’s transparency article. It sits outside the high-risk rules and applies to many ordinary systems. It has four duties that matter here:

  • Article 50(1): providers must design AI systems intended to interact directly with people so that those people are informed they are interacting with an AI system, unless that is obvious to a reasonably well-informed, observant and circumspect person.
  • Article 50(2): providers of systems that generate synthetic audio, image, video or text must ensure the outputs are marked in a machine-readable format and detectable as artificially generated or manipulated. There are exceptions, including systems that perform an assistive function for standard editing or do not substantially alter the input.
  • Article 50(3): deployers of emotion recognition or biometric categorisation systems must inform the people exposed to them.
  • Article 50(4): deployers must disclose deepfakes, and AI-generated text published to inform the public on matters of public interest, unless a person has reviewed it and someone holds editorial responsibility.

Article 50(5) sets the form: the information must be given in a clear and distinguishable manner, at the latest at the time of the first interaction or exposure, and must meet applicable accessibility requirements.

Article 50(6) adds that these duties do not replace other transparency rules in EU or national law. For insurers, that includes the GDPR and the information rules already in insurance distribution law.

The Commission has published guidelines on the transparency obligations for providers and deployers of AI systems, and they are the reference point national authorities will read alongside the text.

Provider or deployer: who carries the duty

The AI Act gives different duties to the provider and the deployer, so the first question is which one you are.

Under Article 3(3), a provider is anyone who develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark. Under Article 3(4), a deployer is anyone who uses an AI system under its authority, other than for purely personal use.

In practice:

  • If you build a claims assistant in house, or commission an engineering partner to build it and run it under your brand, you are its provider and its deployer. Article 50(1) and 50(2) are your duties.
  • If you license a vendor’s chatbot and configure it, the vendor is usually the provider and you are the deployer. The design duty in Article 50(1) sits with the vendor, but your customers will judge you, so check the vendor meets it and put that in the contract.
  • If you substantially change a vendor’s system or put it out under your own name, look again. You may have become the provider.

Illustrative: an MGA licenses a white-label chat widget, connects it to its claims system through an API, and presents it as “Ana, your claims assistant” on its own website, under its own brand. It should not assume the vendor carries Article 50(1). It should map who is the provider for this deployment and agree in writing who designs and maintains the disclosure.

Where Article 50 touches a claims flow

A typical claims journey has several points where an AI system meets the customer.

First notice of loss by chat or voice

A chatbot or voice agent that takes a first notice of loss interacts directly with the claimant. Article 50(1) applies. The “obvious” exception is narrow, and a friendly name with a human-style avatar points the other way. Do not rely on it.

A voice agent that sounds natural is the clearest case for an explicit notice: claimants on a phone line have no visual cue at all.

Status updates and follow-up questions

An agent that answers “where is my claim?” in a portal, by message or by email is still interacting directly with a person. The disclosure belongs in that channel too, not only on the website.

Letters and emails drafted by a model

A system that generates claim decision letters, requests for documents or settlement explanations produces synthetic text. Article 50(2) asks the provider to mark that output in a machine-readable way, unless an exception applies, for example a tool that only helps a handler with standard editing.

For generative systems already on the market before 2 August 2026, the marking duty applies from 2 December 2026. New systems have to meet it now.

Emotion and stress detection on calls

Some call analytics tools score a caller’s emotional state. If you deploy one, Article 50(3) requires you to inform the people exposed to it. Check also whether the use is allowed at all: Article 5(1)(f) bans emotion recognition in the workplace and in education, which matters if the same tool also scores your own handlers.

Pricing and underwriting

Risk assessment and pricing of individuals in life and health insurance is a high-risk use under Annex III, point 5(c). Those systems carry the full high-risk requirements from 2 December 2027, and deployers making or assisting decisions about people must also tell them under Article 26(11). Claims handling is not listed in Annex III point 5, but Article 50 applies to a claims agent whether or not it is high-risk.

What good disclosure looks like

Article 50(5) asks for a notice that is clear, distinguishable and given no later than the first interaction. A workable pattern:

  1. Say it first. The opening message or greeting states that the customer is talking to an AI system. Illustrative wording: “You are chatting with our AI claims assistant. A member of our claims team reviews every decision.”
  2. Keep it visible. A persistent label in the chat header, not only in the first bubble that scrolls away.
  3. Repeat it per channel. Voice, chat, messaging and email each get their own notice, because a claimant may only ever use one.
  4. Offer a person. Not an Article 50 requirement as such, but it lowers complaints and supports the GDPR rules on automated decisions below.
  5. Make it accessible. Screen readers must announce the label, and voice notices must be spoken, not only shown in an app.
  6. Test it. Add a check to your release pipeline that fails if the disclosure string disappears from any channel.

For generated letters, work with the provider of the model or platform on the marking technique. The Commission’s guidelines and the code of practice on transparency of AI-generated content describe the options, such as metadata and watermarks. Keep a record of which technique each channel uses.

The GDPR and insurance rules still apply

Article 50 adds to the GDPR, it does not replace it.

  • Articles 13 and 14 GDPR require you to tell people how their personal data is processed. Where Article 22 applies, you must also give meaningful information about the logic involved in automated decision-making.
  • Article 22 GDPR gives people the right not to be subject to a decision based solely on automated processing that produces legal or similarly significant effects. A claim rejection decided by an agent with no meaningful human review is likely to be the kind of decision this covers. A person signing off on adverse outcomes is the simplest safe design.

EIOPA’s Opinion on AI governance and risk management explains how existing insurance legislation applies to AI systems that are not high-risk or prohibited, which includes most claims assistants. Read it alongside Article 50 when you set your governance for these systems.

What this means for you

  1. List every AI touchpoint in your claims journey: chat, voice, portal, messaging, email and letters. Note which systems you built and which you license.
  2. Decide who is the provider for each one using Article 3(3), and record the answer. Where a vendor is the provider, add the Article 50(1) and 50(2) duties to the contract.
  3. Add a clear AI notice at the first interaction in every channel, keep it visible, and test for it in every release.
  4. Plan the marking of generated text. If your letter-drafting system was on the market before 2 August 2026, the deadline is 2 December 2026. Agree the technique with your model or platform provider now.
  5. Keep a person on adverse decisions so the claims flow meets Article 22 GDPR, and update your privacy notice to describe the agent.
  6. Check for emotion analytics on calls. Inform people where it is used, and confirm nothing scores your own staff.
  7. Run the free AI Act checker for each system to see which obligations apply and from when.

If you want help designing or building a compliant claims agent, see how we work with insurers and MGAs.

Sources

This article is general information, not legal advice.

All insights

Want this built for your team?

One workflow is enough to start. We reply within 24 hours on working days.

Start a projecthello@hasorix.com