AI-written marketing email needs no label, but AI product imagery might
Article 50 of the AI Act applies from 2 August 2026. A human-reviewed, AI-drafted marketing email does not need a label: it fails the public-interest trigger, and it is separately saved by the human-review exception. Photorealistic AI-generated imagery is the real exposure, because the deepfake limb has no human-review exception.
Who this applies to
Deployers sending into the EU. Note the split: the obligation to machine-mark generated output falls on the model provider, not on you. Your exposure is the deployer limb in Article 50(4).
What to do
- Name an accountable human reviewer per send and record it. The Commission excludes superficial, solely formal or procedural checks, so spell-checking does not qualify.
- Treat photorealistic AI product imagery as disclosable. This is the limb with no review exception.
- Look hard at advertorials on health or finance topics styled as editorial. Health is explicitly a public-interest topic, and that combination is the riskiest artefact an email team ships.
- Do not run fully unreviewed autonomous sends into the EU. That forfeits the exception exactly when it matters.
What is exempt
Text that has undergone genuine human review with editorial responsibility, where a named person can approve, alter or reject on substantive grounds. Ordinary marketing also falls outside the public-interest category entirely.
What happens if you do not
Penalties for Article 50 breaches sit in the 15,000,000 euro or 3 percent of worldwide turnover tier. The obligation is one day old at the time of writing, so there is no enforcement record at all yet. The high-risk chapter was deferred to December 2027, but Article 50 was not deferred.
Sources
History of this page
- Status moved to In force as of 2 August 2026.
- Added after the Digital Omnibus confirmed Article 50 was not deferred.