Whether you must tell customers that your text replies come from AI depends on where they are. In the EU, the answer has been yes since 2 August 2026, unless it is already obvious. In the US there is no federal AI disclosure law, but California and Utah have their own rules and the FTC treats deception as unlawful anywhere. The simplest way to stay on the right side of all of them is to say it is an AI in the first message.
This guide covers businesses using an AI agent to send or answer texts. For the separate question of consent to text at all, see our TCPA rules for AI texting.
What does the EU AI Act require for AI text messages?
Article 50 of the EU AI Act sets transparency duties that have applied since 2 August 2026. The part that matters for a texting agent is Article 50(1): providers must design AI systems that interact directly with people so that those people are informed they are interacting with an AI system, unless that is obvious from the circumstances to a reasonably well-informed person.
An SMS or WhatsApp reply rarely makes it obvious. A message that reads like a human employee wrote it is exactly the case the rule targets.
The other parts of Article 50 are less central to a texting agent:
- 50(2) requires machine-readable marking of AI-generated output. Under the Digital Omnibus (Regulation (EU) 2026/1744, in force 27 July 2026), generative systems already on the market have until 2 December 2026 to comply with this marking duty.
- 50(3) covers emotion recognition and biometric categorisation.
- 50(4) covers deepfakes and AI-generated text published to inform the public, with an exception where a human has reviewed and edited it.
- Utah AI Policy Act takes effect
- Utah amendments narrow the disclosure duties
- EU Digital Omnibus enters into force
- EU AI Act Article 50 transparency duties apply
- Deadline for existing generative systems to meet Art. 50(2) marking
If you are outside the EU but text customers who are in it, the obligation can still reach those conversations. Our sister site covers what the AI Act means for people and businesses in the UK.
Is there a US federal law on AI disclosure?
No federal statute requires an AI texting agent to identify itself, as of November 2026. That does not leave a business free to pretend a bot is a person. Section 5 of the FTC Act bars deceptive practices, and the FTC has pursued deceptive AI claims under its Operation AI Comply sweep. A bot that falsely says "I'm Sarah from the front desk" in a way that misleads customers about something that matters to their decisions invites that kind of scrutiny.
The TCPA rules on consent and opt-out apply to AI-sent texts just as they do to any automated text. They are about whether you can text someone, not whether you say it is AI, but a clear disclosure in the first message is a natural place to put your opt-out wording too.
What does California's bot law say?
California's bot disclosure law (SB 1001, Business and Professions Code section 17940 onwards) has been operative since 1 July 2019. According to Cooley's summary, it makes it unlawful to use a bot to communicate or interact online with a person in California with the intent to mislead them about the bot's artificial identity, in order to:
- incentivise a purchase or sale of goods or services in a commercial transaction, or
- influence a vote in an election.
The law defines a bot as an automated online account where all or substantially all of the actions or posts are not the result of a person. A disclosure that is clear, conspicuous and reasonably designed to inform people that they are dealing with a bot means there is no violation.
Two points are unsettled. First, the law speaks of online communication, and we did not find an authoritative answer on whether an SMS conversation counts. Second, reports describe proposed amendments, and we could not confirm their status. For a sales or lead-qualification bot texting Californians, the practical answer is the same either way: disclose. Our AI lead qualification guide covers bots that are most exposed to the "incentivise a purchase" wording.
What does Utah's AI Policy Act require?
Utah's Artificial Intelligence Policy Act took effect on 1 May 2024 and was narrowed by two bills effective 7 May 2025, according to Davis Polk. Its repeal date was pushed back to July 2027. As amended, it works in two tiers:
| Situation | What Utah requires |
|---|---|
| A consumer clearly and unambiguously asks whether they are dealing with AI | Disclose that they are |
| High-risk interaction in a state-regulated occupation (sensitive personal data plus advice relied on for significant financial, legal, medical or mental health decisions) | Disclose prominently up front |
| Routine messages such as appointment confirmations and reminders | The narrowed definition of an AI system excludes these, per the law-firm summaries |
There is a safe harbour for a clear and conspicuous disclosure at the start of the interaction and throughout it. Fines are up to $2,500 per violation, and there is no private right of action.
For a texting agent, the first tier means your bot must never deny being an AI when asked. That is a prompt-level rule you can enforce. The second tier matters for clinics, financial advisers and law firms.
Does the UK require AI disclosure in texts?
The EU AI Act does not apply in the UK. There is no UK statute equivalent to Article 50(1) for chatbots, as of November 2026. UK businesses still face PECR rules on marketing texts and UK GDPR transparency duties when personal data is processed. Our PECR guide for AI text messages covers the marketing side. A UK business that texts customers in the EU should treat those conversations as within the AI Act's reach.
How should a texting agent disclose that it is AI?
A single honest line at the start of the conversation meets the strictest rule above and costs almost nothing.
- Say it in the first messageFor example, "Hi, this is Acme's AI assistant."
- Offer a humanAdd a way to reach a person, such as "Reply HUMAN to talk to our team."
- Keep opt-out wordingInclude "Reply STOP to opt out" where your messages require it.
- Never deny being AIInstruct the model to confirm it is an AI whenever asked, and test that rule.
- Repeat after a gapIf a conversation resumes after days or weeks, reintroduce the assistant.
Some practical points:
- Do not give the bot a human persona with a human name and backstory. A named assistant is fine if the message says it is an AI assistant.
- Enforce the rule in code as well as the prompt. Prompts can be overridden, including by prompt injection through incoming messages. A fixed, code-generated first message is more reliable than relying on the model to introduce itself.
- Watch segment length. A disclosure plus opt-out line adds characters. Keep it short so the first reply still fits in one SMS segment.
Pros
- Meets EU Article 50(1) and Utah's safe harbour
- Removes the California "intent to mislead" risk
- Builds trust and sets expectations for errors
Cons
- Adds a few characters to the first message
- Some customers may ask for a human more often
Bottom line
Disclose AI in the first text. Since 2 August 2026 it has been required in the EU unless obvious. In the US, California bars misleading bot identity to drive sales, Utah requires an honest answer when asked and up-front disclosure for some regulated advice, and the FTC Act bars deception everywhere. The UK has no direct equivalent, but EU customers bring the AI Act with them. A one-line introduction plus a route to a human handles all of it. For the wider compliance picture, start with our SMS AI agents for business guide and the safeaiagents.com overview of US AI agent rules.


