AI email replies and Article 50: when an email needs an AI disclosure

An AI that writes and sends email replies on its own is considered to be directly interacting with customers under Article 50 of the AI Act. This means that such AI-generated replies must disclose that they were produced by an AI rather than a human. Rule-based auto-responders like out-of-office replies or standard "we received your request" responses do not qualify as AI systems and are therefore not subject to the disclosure requirement.

In cases where an AI drafts a reply but a human reviews and edits it before sending, the review must be a genuine process where the human is actively engaging with the AI output to ensure accuracy and appropriateness. If this review is merely a formality with little actual human oversight, the reply can be treated as AI-sent.

For email workflows involving AI agents that autonomously write and send follow-ups to prospects, the AI disclosure must be included in each AI-generated reply. The sender name used for AI-sent replies should not imply human authorship, such as using a generic "Brightline Support" rather than "Emma from Brightline". In outbound sales and marketing sequences, the same AI disclosure rule applies to all AI-created communications.

The disclosure should appear prominently at the beginning of the email body, rather than in the footer, to ensure it is read first. It must be clear, distinguishable, and appear at the first interaction with the customer. For AI-drafted replies that are then sent by staff, the review should be thorough and documented, showing the human has actually engaged with the AI output. If staff cannot meet these standards, the replies should be treated as AI-sent and the disclosure added.