What does Commission guidelines on AI Act Article 50 transparency say about AI in banking?

What are the key points of Commission guidelines on AI Act Article 50 transparency?

• Status: adopted 20 July 2026 as C(2026) 5054; the Guidelines are non-binding (paragraph 5) and issued under Article 96(1)(d) of the AI Act. The Commission published the final Code of Practice on marking and labelling AI-generated content on 10 June 2026 as a voluntary compliance route.
• Article 50(1): providers must design AI systems intended to interact directly with natural persons so that those persons are informed; exceptions where the artificial origin is obvious or where the system is authorised by law for law enforcement. The Guidelines say helpdesk chatbots embedded in platforms and similar tools users may perceive as human do not satisfy the obviousness exception.
• Direct interaction (Section 3.1): excludes indirect cases such as customer service representatives who use AI assistance tools, but the mere possibility of human review is not a reason to avoid the duty.
• Article 50(2): providers of systems generating synthetic audio, image, video or text must mark outputs in a machine-readable format and make them detectable with effective, interoperable, robust and reliable technical solutions; exceptions for assistive editing functions that do not substantially alter input.
• Article 50(3) and (4): deployers must inform people exposed to emotion recognition or biometric categorisation systems, disclose deep fakes, and label AI-generated text published to inform the public on matters of public interest unless it has human review or editorial responsibility.
• Article 50(5) (Section 7): information must be given clearly and distinguishably at the first interaction or exposure and meet accessibility requirements.
• Enforcement (Section 8): fines up to EUR 15 000 000 or, for undertakings, up to 3% of worldwide annual turnover, whichever is higher (paragraph 152); lower of the two for SMEs.
• Entry into application (paragraph 153): Article 50 applies from 2 August 2026 to all in-scope systems, including those already on the market; the AI Omnibus grandfathering is limited to the Article 50(2) marking and detection duty, with a transition to 2 December 2026. Partly interactive and partly generative systems must still meet the Article 50(1) disclosure from 2 August 2026.

What did Commission guidelines on AI Act Article 50 transparency change for banks?

Article 50 is the part of the AI Act that reaches almost every bank using a customer-facing assistant, whatever its risk class. The Guidelines clarify that customer-service chatbots generally must disclose that they are AI, and that an obvious-interaction exception is narrow. The Digital Omnibus left the 2 August 2026 date untouched for the chatbot disclosure duty, so that obligation is already live for banks.

What do the Commission's Article 50 guidelines require of banks using chatbots and generative AI?

The Commission's Article 50 Guidelines (C(2026) 5054, 20 July 2026) explain that, from 2 August 2026, the provider of an AI system that interacts directly with people must inform them they are dealing with AI unless that is obvious, that providers of generative systems must mark outputs in a machine-readable way, and that deployers must disclose deep fakes and certain AI-generated public-interest text. A bank that deploys an AI customer assistant should ensure the disclosure is given at the first interaction, in an accessible way. Breaches can be fined up to EUR 15 million or 3% of worldwide turnover, and systems already on the market have until 2 December 2026 only for the marking duty.

Article 50 sits in the AI Act's transparency tier, separate from the high-risk regime for credit scoring, which the Digital Omnibus moved to 2 December 2027. It therefore reaches customer-facing tools such as banking assistants, voice bots and AI-generated marketing material even when the underlying system is not high-risk.

The Guidelines distinguish providers, who must build disclosure and marking into the system, from deployers, who carry the disclosure duties for deep fakes and public-interest text. A bank that buys a chatbot from a vendor will usually be a deployer, but one that builds or substantially modifies a customer-facing system for its own use is likely to be its provider and so responsible for the Article 50(1) design duty.

Compliance can be supported by the voluntary Code of Practice on marking and labelling AI-generated content, published on 10 June 2026, and enforcement is by national market surveillance authorities, with the AI Office competent for systems built on general-purpose models from the same provider.

WHAT THIS MEANS IN PRACTICE

• Inventory every customer-facing AI interface (chat, voice, email assistants) and confirm each tells the user at the start that it is an AI system, unless obviousness can be defended for that audience.
• Do not rely on the possibility of human review to avoid disclosure; the Guidelines say that does not remove the Article 50(1) duty.
• For generative tools, require vendors to evidence machine-readable marking and plan for the 2 December 2026 transition date for legacy systems.
• Check marketing, investor and public communications for AI-generated or manipulated images, audio or video that may be deep fakes and for AI-written public-interest text without human review.
• Assign ownership of Article 50 compliance between product, compliance and vendor management, and record who is provider and who is deployer for each tool.