More and more business email is written with AI, from Gemini drafts in Gmail to AI agents that run whole sales and support conversations. Since August 2, 2026, the EU AI Act’s transparency rules apply, and many teams are asking whether every email now needs a “written by AI” label. For most emails the answer is no. If an AI agent emails people on your company’s behalf, though, the rules do apply, and where you put the disclosure matters.
Quick answer: An email that a person writes with help from AI doesn’t need an AI label under the EU AI Act, and neither does ordinary marketing copy. A disclosure is needed when an AI agent emails people on its own: recipients must be told, clearly and no later than the first message, that they’re dealing with AI and who it acts for. Realistic AI-generated images of real people, places or events also need a label. These rules have applied since August 2, 2026.
This is general information, not legal or compliance advice. How the AI Act applies depends on your setup, your AI vendors, and where your recipients are. Confirm your approach with your legal or compliance team before you roll anything out.
What Changed on August 2, 2026
The EU AI Act (Regulation (EU) 2024/1689) has been rolling out in stages since 2024. Its transparency rules, in Article 50, started applying on August 2, 2026. On July 20, 2026, the European Commission published final guidelines explaining how those rules work in practice.
Article 50 sets out four transparency duties. Two of them matter for email:
- AI that interacts with people (Article 50(1)). AI systems that interact directly with people must tell them they’re dealing with AI, unless that’s obvious. The Commission’s guidelines treat AI agents as covered, and agents must disclose both that they’re AI and who they’re acting for.
- Deep fakes and public-interest text (Article 50(4)). Companies that use AI to create realistic fake images, audio or video, or AI-written text published to inform the public on matters of public interest, must disclose it. The text rule doesn’t apply when a person reviews the content and takes editorial responsibility for it.
The other two duties cover AI providers marking their output in a machine-readable way, and systems that recognize emotions or sort people by biometric data. Those rarely affect everyday email.
Three more details matter for anyone sending email:
- Timing and format. The information has to be clear and distinguishable, given at the latest at the first interaction or exposure, and accessible (Article 50(5)).
- Reach. The Act can apply to companies outside the EU when their AI output is used in the EU (Article 2(1)(c)). A US company whose AI agent emails prospects in Germany can be in scope.
- Penalties. Breaking the transparency rules can lead to fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher. For small and medium-sized companies, the lower of the two applies (Article 99).
Which Emails Need an AI Disclosure (and Which Don’t)
Most of the confusion comes from treating every “AI email” the same way. The AI Act looks at who is actually talking to the recipient and what the content is.
| Email scenario | AI disclosure required? | Why |
|---|---|---|
| A person drafts an email with Gemini or Copilot, edits it, and sends it | No | The person is the sender and the AI is a writing tool. Business correspondence isn’t a publication meant to inform the public. |
| Marketing emails or newsletters with AI-written copy | Usually no | Promotional copy isn’t text published to inform the public on matters of public interest. |
| An AI agent (for example an AI SDR or support agent) sends emails and answers replies on its own | Yes, treat it as required | The agent interacts directly with people, so they need to know it’s AI and who it acts for. |
| News-style content on public-interest topics (health, finance, politics) written by AI and sent without real editorial review | Likely yes | Article 50(4) applies unless a person reviews it and holds editorial responsibility. |
| An email with a realistic AI-generated image or video of real people, places or events | Yes, for that image or video | Deep fakes must be disclosed. |
An agent that writes to people and handles their replies is interacting with them directly, so the safe approach is to treat its emails as covered.
Keep two more things in mind. Other rules still apply to every email, whoever wrote it: GDPR, consumer protection law, and industry rules like FINRA or FERPA. And the editorial exception is narrow. Law firms reviewing the guidelines, such as Paul, Weiss, note that spell-checking or a quick sign-off doesn’t count as editorial review.
What a Compliant AI Disclosure Includes
When a disclosure is required, the bar is fairly high. Based on Article 50 and the Commission’s guidance, a good AI disclosure in an email:
- Says “AI” in plain words. A vague label like “assistant” isn’t enough on its own, and neither is a mention buried in your terms and conditions.
- Names who the AI acts for. An agent should make clear it’s acting on behalf of your company.
- Appears where people see it first. Article 50(5) asks for the information at the latest at the first interaction or exposure. In an email, the safest place for it is the top of the message.
- Shows up on every agent email. Replies get forwarded and new people get added in CC, so any message can be someone’s first contact with the agent.
- Is accessible. Use live text, not an image, so screen readers and the plain-text version of the email show it too.
- Gives a way to reach a person. The Act doesn’t spell this out, but it’s good practice.
Example AI Disclosure Templates
Adapt these with your legal team and replace the bracketed text.
1. Notice at the top of an AI agent’s email (sales or outreach)
This email was written and sent by an AI agent on behalf of [Company]. To talk to a person on our team, email [name@company.com].
2. Notice at the top of an AI support reply
This reply was generated by [Company]’s AI support agent. If it doesn’t solve your issue, reply “person” and someone from our support team will take over.
3. Signature block for an AI agent mailbox
[Agent name]
AI agent, [Company]
This is an AI system acting on behalf of [Company].
To reach our team: [email] | [phone]
4. Optional line for people who draft with AI
Drafted with AI assistance and reviewed by [Name].
The AI Act doesn’t require this line on emails a person writes. Some companies add it anyway as part of an AI transparency policy. If you do, use the same wording everywhere so it means the same thing on every email.
5. Label for an AI-generated image
Image created with AI.
Place it right next to the image and say the same thing in the image’s alt text.
For other disclaimer types, such as confidentiality notices, see our email disclaimer templates.
Where to Put the Disclosure: Top of the Email or the Signature?
The email signature feels like the natural home for a disclaimer, and for confidentiality notices it usually is. For AI agents, a signature line on its own is risky for two reasons.
People read from the top. Article 50(5) asks for the disclosure at the latest at the first interaction or exposure. A line at the bottom of an email, below the message and sometimes below a long quoted thread, may be the last thing a reader sees. Put the main notice at the top of every agent email and repeat it in the signature as a backup.
Agents can skip the signature entirely. Signatures set in Gmail are added when someone writes the email in the Gmail web interface, as Google’s own documentation explains. Many AI agents send email through an API instead of an email app, so they may never pick up the signature at all. Build the notice into the agent’s message template, then check a real sent email to confirm it’s there.
If your agent sends from its own mailbox in Google Workspace or Microsoft 365, the admin console gives you a server-side backup:
- Google Workspace can add a standard footer to outgoing emails for everyone in an organizational unit. Google adds it on its servers, so it also covers mail sent from other clients such as Outlook or Apple Mail. Put the agent’s account in its own organizational unit so only its emails get the footer. The footer sits at the bottom of the message, and changes can take up to 24 hours.
- Exchange Online mail flow rules can add a disclaimer to messages, scoped to the agent’s mailbox. You can choose to prepend it, which places it at the top. If a message is encrypted, the rule can’t add the text, so you set a fallback action.
Test both with your agent before you rely on them.
Common AI Disclosure Mistakes
- Using vague wording. “Your assistant Ava” doesn’t tell anyone they’re dealing with AI. Say “AI agent” or “AI system.”
- Disclosing only in the footer. A line at the bottom of a long thread is easy to miss. Lead with the notice.
- Assuming the signature is there. If the agent sends through an API, check a real sent email.
- Putting the disclosure in an image. Many email clients block images by default, and screen readers can’t read text inside them. Use live text.
- Letting every team write its own version. When sales, support and marketing each use different wording, recipients get mixed signals and your compliance team can’t tell which version is approved.
- Labeling everything “AI-generated.” Tagging every human-written email isn’t required and makes the label easier to ignore. Save it for the cases that need it.
- Forgetting AI images. A realistic AI-generated photo of a real person or place in a campaign email needs its own label.
Managing AI Disclosures Manually vs. With Software
| Manual copy and paste | Native footers and mail flow rules | BulkSignature | |
|---|---|---|---|
| Where the text appears | In each person’s own signature | In a separate block added to the message | Inside each signature, as part of your approved template |
| Who gets it | Whoever remembers to paste it | Everyone in an organizational unit (Google) or messages that match a rule (Exchange) | The departments, organizational units, groups or users you choose |
| Changing the wording | Edit every mailbox by hand | One change in the admin console | One change, pushed to every signature that uses it |
| Look and consistency | Varies from person to person | Plain block, separate from the signature | Matches your brand template on every email |
| Best for | Very small teams | AI agent mailboxes and catch-all notices | Teams of people across Google Workspace and Microsoft 365 |
When You Need Email Signature Software
If you have one AI agent and one notice to add, the agent’s own template plus a server-side footer may be all you need. Signature software earns its place when the AI disclosure is part of a wider legal and compliance setup:
- Several teams need different wording, for example support versus sales.
- You already manage confidentiality notices, GDPR lines or regulated disclaimers in signatures.
- Your legal team wants one approved version and a record of every change.
- Your people work across Google Workspace and Microsoft 365, or across several offices.
- The wording is likely to change as the guidance and your AI tools evolve.
How to Add AI Disclosures to Your Company’s Emails
- Map where AI touches email. List drafting tools like Gemini and Copilot, AI agents that send email, AI auto-replies, and AI-generated images in campaigns.
- Sort each flow using the table above. Mark which ones are AI interacting with people, which involve deep fakes or public-interest content, and which need no label.
- Write approved wording with legal. Keep it short, say “AI,” name your company, and give a way to reach a person.
- Put the notice at the top of every agent email. Set it in the agent’s message template or settings, then check real sent emails, including replies and follow-ups.
- Add a backup line in the signature or a server-side footer. Give the agent’s mailbox its own signature template, or its own organizational unit with a footer or mail flow rule.
- Keep records. Note which AI tools you use, what each disclosure says, and who reviews content where you rely on the editorial exception.
- Review it regularly. Recheck the wording when you add an AI tool, change vendors, or new guidance comes out.
AI Email Disclosure Checklist
- Every AI agent that emails people says it’s AI and who it acts for
- The notice sits at the top of each agent email, including replies and follow-ups
- The wording says “AI” plainly and avoids vague labels like “assistant”
- The disclosure is live text and appears in the plain-text version of the email
- Agent mailboxes have a backup line in the signature or a server-side footer
- A real sent email has been checked for every agent and channel
- Realistic AI-generated images and video are labeled
- Human review is documented wherever you rely on the editorial exception
- Your AI vendors’ disclosure settings are confirmed and switched on
- One approved wording is used across teams, with a record of changes
Keeping AI Disclosures Consistent With BulkSignature
BulkSignature helps IT and compliance teams manage the signature side of AI transparency from one place across Google Workspace and Microsoft 365. With the Compliance Center, you can:
- Add an approved disclosure or legal footer to every signature in a department, organizational unit or group, without asking anyone to paste text.
- Change the wording once and push it to every signature that uses it.
- See who changed what and when, which helps when legal asks for a record.
- Keep AI disclosures next to your other required lines, such as confidentiality notices, GDPR information and industry disclaimers.
For AI agents, set the top-of-email notice in the agent’s own settings. BulkSignature keeps the signatures of the people and teams working alongside those agents consistent and on-brand.
Frequently Asked Questions About AI Email Disclaimers
Does the EU AI Act require a disclaimer on every AI-written email?
No. Emails that a person writes with help from AI, then reviews and sends, don’t need an AI label under the Act. Ordinary marketing copy usually doesn’t either. The disclosure duties focus on AI that interacts with people directly, deep fakes, and AI-written text published to inform the public on matters of public interest.
Do emails from AI sales or support agents need a disclosure?
Yes, if the agent interacts with people on its own. Recipients need to know they’re dealing with AI and who the agent acts for, at the latest when they first interact with it. The legal duty to build this in sits with the provider of the AI system, which is usually you if you built the agent yourself and the vendor if you use theirs. Either way, the disclosure appears in your emails, so check that it’s switched on and clearly worded.
Is an AI disclaimer in the email signature enough?
Often not on its own. The disclosure has to reach people at the latest at the first interaction, and a signature sits at the bottom of the email, sometimes below a long quoted thread. Many agents also send through APIs that skip the signature. Put the notice at the top of agent emails and use the signature as a backup.
Does the EU AI Act apply to companies outside the EU?
It can. Article 2 covers providers and deployers outside the EU when the output of their AI system is used in the EU. If your AI agent emails people in the EU, plan as if the transparency rules apply. Other countries and some US states have their own AI or bot disclosure rules, so check where your recipients are.
What should an AI disclosure in an email say?
Keep it short and plain. Say that the email was written or sent by AI, name the company it acts for, and give a way to reach a person. For example: “This email was written and sent by an AI agent on behalf of [Company]. To talk to a person on our team, email [address].”
Do AI-generated images in marketing emails need a label?
Some do. A realistic AI-generated image or video that shows real people, places or events and could pass as real counts as a deep fake, and it must be disclosed. Clearly illustrated or stylized artwork usually doesn’t fall into that category. When in doubt, add a short caption such as “Image created with AI.”
What are the penalties for missing an AI disclosure?
Breaches of the transparency rules can lead to fines of up to €15 million or 3% of a company’s worldwide annual turnover, whichever is higher. For small and medium-sized companies, the lower of the two amounts applies.
Sources: EU AI Act, Regulation (EU) 2024/1689 · European Commission guidelines on Article 50 · Paul, Weiss: EU finalises transparency rules for AI-generated content · Gmail API: send-as signatures · Google Workspace: add a standard footer to outgoing emails · Microsoft Learn: disclaimers in Exchange Online
