Consent and Suppression Rules an Outreach Agent Must Follow by Region

In an autonomous outreach workflow, consent, suppression lists and regional privacy compliance belong in a gate the agent cannot argue with. Our position is that an agent should never reason its way to "this person is fine to contact". It reads three recorded facts (which rule set applies, what basis exists, whether the address is suppressed) and holds the send when any of them is missing. Leave the model's judgment to the drafting, which our primer on agentic AI covers, and make permission to send a lookup built from regulators' published rules. What follows is an operating design, so have counsel confirm how each law applies to your company.

Three facts the agent reads before every send

The check runs at send time for every step of a sequence, because a contact who was clear at enrollment may have opted out an hour ago. In the design we propose, the agent needs:

1. Region rule. Derived from recorded evidence such as billing country, the state given on a form, or the company's registered office. An unknown region resolves to the strictest rule you operate under, or to a hold.
2. Basis. A stored reason that the rule accepts: an opt-out regime, express consent with its record, or an implied basis with its start date.
3. Suppression status. A read of the suppression list taken seconds before the send, with the list version recorded.

This is a design to build or configure, so confirm in your sequencing tool's documentation that it can read suppression at send time and block on the result.

What does the FTC require for a CAN-SPAM opt-out?

The official FTC guidance on CAN-SPAM opt-out requirements is short enough to encode in full. Per the FTC compliance guide (August 2023, edited January 2024), an opt-out mechanism has to keep processing requests for at least 30 days after the send, and a request must be honored within 10 business days. The sender cannot charge a fee, ask for personally identifying information beyond an email address, or require more than a reply email or a visit to a single web page. A preference menu is allowed when it includes an option to stop all marketing messages.

None of the eight main requirements in the guide asks for consent before the first message. They cover accurate headers, honest subject lines, disclosure that the message is an ad, a valid physical postal address, the opt-out, and oversight of anyone sending on your behalf. Two statements matter most to agent builders: "The law makes no exception for business-to-business email", and each separate email in violation is subject to penalties of up to $53,088. After an opt-out the address cannot be sold or transferred, except to a company hired to help you comply. The guide also tells senders: "Make sure your spam filter doesn't block these opt-out requests." Since a reply email is one of the two routes the guide names, we treat a reply that says "take me off your list" as an opt-out even when nobody clicks the unsubscribe link, and the agent's reply classifier writes it to the suppression list.

What California's opt-out of sale or sharing covers

California's CCPA gives a consumer the right to opt out of the sale or sharing of personal information, and the official privacy rights page from the California Attorney General (updated August 28, 2026) is where direct marketing teams should start. Sharing, on that page, "refers specifically to sharing for cross-context behavioral advertising", and we found no separate California right there to opt out of receiving email; CAN-SPAM supplies that one. In our design the CCPA flag gates what the agent does with a contact's data, such as pushing a prospect list into an ad platform audience or passing records to a partner.

The page applies the law to for-profit businesses that do business in California and meet any of three tests: gross annual revenue over $25 million; buying, selling or sharing the personal information of 100,000 or more California residents or households; or deriving 50% or more of annual revenue from selling that information. The California Privacy Protection Agency FAQ gives the current figures as gross annual revenue of $26.625 million or more, effective January 1, 2025, and 50% or more of annual revenue from selling or sharing that information. The exemption for personal information reflecting business-to-business transactions expired on December 31, 2022, so a covered sender should treat a prospect's work details as in scope. Three rules translate directly into agent behavior:

• A business must respond to an opt-out request as soon as feasibly possible, within a maximum of 15 business days.
• A Global Privacy Control signal must be honored as a valid opt-out request.
• "Businesses must wait at least 12 months before asking you to opt back in to the sale or sharing of your personal information", so each record carries a re-ask date.

Bought data adds a second clock. The California Privacy Protection Agency says that under the Delete Act, beginning August 1, 2026, data brokers must access the "accessible deletion mechanism" at least once every 45 days and process consumer deletion requests, subject to limited exceptions. We propose stamping every enriched record with its vendor and pull date, and refreshing anything older than 45 days before a send.

Sensitive personal information stays out of the prompt

California privacy law pairs the CCPA opt-out of sale and sharing with a right to limit the use and disclosure of sensitive personal information, and the official definition of that category is worth reading before you choose enrichment fields. The Attorney General's page lists certain government identifiers, account log-ins and card numbers with their credentials, precise geolocation, genetic data, health information, racial or ethnic origin, union membership and "contents of mail, email, and text messages", among others.

Our rule is stricter than the page requires and cheaper to run: no field in those categories reaches the agent as a personalization input, in any region. Filtering by category at the data layer is simpler than teaching a model when a limit request applies.

How the check changes in the UK, Canada and the EU

The ICO says the soft opt-in "does not apply to prospective customers or new contacts (eg from bought-in lists)", that the same rule covers direct messages via social media, and that its guidance is under review following the Data (Use and Access) Act. Canada's anti-spam law, as published on the Justice Laws Website and current to September 21, 2026, says a person who claims consent "has the onus of proving it" and sets the maximum penalty per violation at $10,000,000 for anyone other than an individual. The European Commission says a company using an acquired list must inform individuals "at the latest at the time of the first communication with them", and it points to Article 13 of the ePrivacy Directive for email. We did not open that article for this piece, so in our design EU contacts default to express consent until counsel sets a country rule.

One suppression list, written by everything

Each opt-out deadline quoted above is measured in business days, and we think an agent should plan to use none of them. A check at send time blocks the very next message and leaves the legal window as margin for whatever your systems miss. Our proposed design:

• One list, keyed by normalized address, with domain-level entries for organizations that object.
• Every channel writes to it: unsubscribe links, classified replies, CRM status changes, vendor deletion files and privacy requests.
• The agent adds and never removes. Removal takes a named person, in line with what marketing AI agents can write back to your CRM.
• Imports are screened before enrollment, so a vendor refresh cannot bring an opted-out contact back.
• All agents share it. Two agents with separate copies will eventually email someone one of them suppressed, and shared controls of this kind are the subject of the governance layer for multi-agent stacks.

The log entry for one hypothetical send

The contact and dates below are invented to show the record, and the expiry date applies the six-month inquiry window from the Canada row above.

If the same contact were scheduled for a step on December 10, the check would return a block with the reason "basis expired", and the sequence would end without a person having to notice.

Where people stay involved, and what to measure

A person approves the first send under any new rule set or data source, every removal from suppression, and a weekly sample of replies the classifier marked ambiguous. Those gates sit inside the wider governance operating model for agents.

Measure the gate like any other part of the funnel: sends blocked by reason, the share of contacts with an unknown region, median and maximum time from opt-out to block, sends after an opt-out (the target is zero), and the share of sends with a complete log entry. Blocked contacts shrink the reachable list, so subtract them before you model return; our AI ROI calculator is a further resource for that sum. If you want the gate built into a live sequence, our agentic AI automation team can help.

An agent that asks permission from a table will send fewer messages than one that judges for itself. We would take that trade every time, because each message it does send can be explained in one row.

Sources

• FTC: CAN-SPAM Act: A Compliance Guide for Business (August 2023, edited January 2024)
• California Attorney General: California Consumer Privacy Act (CCPA) (updated August 28, 2026)
• California Privacy Protection Agency: Information for Data Brokers
• California Privacy Protection Agency: Frequently Asked Questions (FAQs)
• ICO: Electronic mail marketing, Guide to PECR
• Justice Laws Website: Canada's anti-spam legislation, S.C. 2010, c. 23 (current to 2026-09-21)
• European Commission: Can data received from a third party be used for marketing?