Does Article 50 ban deploying AI in the EU without a proof mechanism?

No. Article 50 is a transparency duty, not a prohibition. Anyone telling you that deployment without their product is unlawful is describing their commercial interest, not the regulation.

Is a provenance chain mandatory?

No. Point (73) of Commission guidelines C(2026)5054 of 20 July 2026 says a full provenance chain is not required. We keep one because it answers the supervisor's second question, not because the law asks for it.

Who carries the obligation: the platform or the agency?

It depends on the role. A party that puts an AI system on the market under its own name acts as provider; a party that operates it under its own authority acts as deployer. A white-label agency can be both at once, which is exactly why position 4 asks for the allocation to be written down.

Position paper · Article 50

Our position on Article 50: verifiable, not merely declared.

Article 50 applicable since 2 August 2026

Last reviewed: 31 August 2026

Article 50 of the EU AI Act obliges you to tell a person they are dealing with an AI system, and to mark synthetic content as machine-generated. It does not say how you show, afterwards, that this happened. That gap is where our five positions sit, and they are written so that any provider can implement them, not only us.

What Article 50 actually requires

A natural person interacting with an AI system must be informed of that, unless it is obvious from the context.: Regulation (EU) 2024/1689, Art. 50(1)
Synthetic audio, image, video and text must be marked in a machine-readable way as artificially generated or manipulated.: Regulation (EU) 2024/1689, Art. 50(2)
A full provenance chain is not mandatory.: Commission guidelines C(2026)5054, 20 July 2026, point (73)
For SMEs and start-ups the administrative-fine ceiling is the lower of €15 million and 3% of worldwide annual turnover, not the higher.: Regulation (EU) 2024/1689, Art. 99(6); guidelines C(2026)5054, point (152)
Systems already on the market before 2 August 2026 have until 2 December 2026 to satisfy Art. 50(2).: Regulation (EU) 2026/1744 (Digital Omnibus), inserting Art. 111(4)

The positions

Five positions, in the order they depend on each other.

Each one is written to be implementable by any provider of AI systems that speak to people. None of them requires an Audact product, and that is deliberate: a position that only its author can implement is a sales argument wearing a suit.

#
1
Position
Verifiability over declaration
Why
An obligation that can only be self-declared cannot be supervised at scale. Every AI interaction with a natural person should leave a tamper-evident, machine-checkable record covering disclosure, consent, jurisdiction and time.
#
2
Position
Machine-readable evidence
Why
Evidence should be expressible in existing open formats (W3C Verifiable Credentials, C2PA content credentials), so that an authority can read it with standard tooling instead of one vendor dashboard per supplier.
#
3
Position
A policy layer per member state
Why
Article 50 does not stand alone. It sits next to national rules: the Dutch Telecommunicatiewet, requirements at Bundesland level in Germany, CNIL guidance in France. Without a public, machine-readable rule schema every provider maintains its own incompatible interpretation.
#
4
Position
Contractual allocation between provider and deployer
Why
Liability should be allocated in writing, with default rules for the case where the contract is silent. Without that, an agency acting as deployer carries an undefined residual exposure for the behaviour of the platform above it.
#
5
Position
Portable evidence formats
Why
Compliance evidence should be portable, not a lock-in mechanism. A format that only one supplier can read turns a legal obligation into a switching cost, and that is bad for the market and worse for supervision.

Five positions on Article 50
#	Position	Why
1	Verifiability over declaration	An obligation that can only be self-declared cannot be supervised at scale. Every AI interaction with a natural person should leave a tamper-evident, machine-checkable record covering disclosure, consent, jurisdiction and time.
2	Machine-readable evidence	Evidence should be expressible in existing open formats (W3C Verifiable Credentials, C2PA content credentials), so that an authority can read it with standard tooling instead of one vendor dashboard per supplier.
3	A policy layer per member state	Article 50 does not stand alone. It sits next to national rules: the Dutch Telecommunicatiewet, requirements at Bundesland level in Germany, CNIL guidance in France. Without a public, machine-readable rule schema every provider maintains its own incompatible interpretation.
4	Contractual allocation between provider and deployer	Liability should be allocated in writing, with default rules for the case where the contract is silent. Without that, an agency acting as deployer carries an undefined residual exposure for the behaviour of the platform above it.
5	Portable evidence formats	Compliance evidence should be portable, not a lock-in mechanism. A format that only one supplier can read turns a legal obligation into a switching cost, and that is bad for the market and worse for supervision.

Why it matters

Disclosure is easy to claim and hard to show.

Delivering an Article 50 disclosure is a small piece of engineering. Almost every voice platform on the market can do it, and most say so. Showing six months later, for one specific call, in which language the disclosure was delivered and against which policy version, is a different problem, and it is the one that arrives with the letter from the supervisor.

That is the whole of our argument. Not that the law demands proof, because point (73) of the guidelines says it does not. But that a duty which can only be self-declared is a duty nobody can check, and that the format of the answer should be settled before the first inquiry rather than after it.

Questions

Does Article 50 ban deploying AI in the EU without a proof mechanism?: No. Article 50 is a transparency duty, not a prohibition. Anyone telling you that deployment without their product is unlawful is describing their commercial interest, not the regulation.
Is a provenance chain mandatory?: No. Point (73) of Commission guidelines C(2026)5054 of 20 July 2026 says a full provenance chain is not required. We keep one because it answers the supervisor's second question, not because the law asks for it.
Who carries the obligation: the platform or the agency?: It depends on the role. A party that puts an AI system on the market under its own name acts as provider; a party that operates it under its own authority acts as deployer. A white-label agency can be both at once, which is exactly why position 4 asks for the allocation to be written down.

Ask for a technical briefing.

For national supervisory authorities, DPOs and regulated operators: we will walk through the five positions and the mechanism behind them, without a sales deck.

regulators@audact.ai