EU AI Act Article 50 takes effect: AI assistants must disclose their identity

On 5 October 2026, a developer who runs LLM agents in production published a post on DEV Community laying out when Article 50 of the EU AI Act actually takes effect. The central warning is that many teams have latched onto the 2 December 2026 date and assume they have until then to comply. For the most important part of the obligation, the author argues, that assumption is wrong.

The duty to tell a person that they are talking to an AI has already applied since 2 August, according to the post. The 2 December deadline covers only one technical point, and only for systems that are already in service: a machine-readable marking requirement.

The post also addresses who carries the responsibility. In the author's account, a party that builds a system on top of third-party model APIs is itself the provider of that system, and therefore holds the compliance obligation rather than passing it up to the upstream model vendor. The stated maximum penalty is EUR 15 million or 3% of global turnover.

The author is explicit that he is not a lawyer. The write-up comes from his own practical experience as someone operating LLM agents in production, including one agent that sends messages to strangers. Beyond the dates, the scope of the obligation, the question of who counts as the provider and the fine ceiling, the post does not offer further detail.