Five Letters, Two Circuits, One Ledger: Build the Consent Record Your AI Agents Must Read Before They Speak

One consumer typed five letters. Two federal courts, one month apart, read them in opposite directions.

On July 14 the Seventh Circuit held that a text message is not a "telephone call," which means a consumer who replies STOP to a marketing text and gets more texts anyway has no private lawsuit under the Do Not Call provision. On August 11 a federal judge in the Southern District of Florida held the opposite kind of thing: that same STOP, sent to a text, may have revoked consent for phone calls too.

Both are 2026. Both are federal. Neither is wrong. And the operational lesson underneath them has nothing to do with either holding.

The Two Rulings, Side by Side

In Steidinger v. Blackstone Medical Services, the Seventh Circuit went to the 1991 dictionary. Congress wrote a private remedy in section 227(c)(5) for repeated "telephone call[s]," and in 1991 a call meant a voice. Venable's Shahin Rothermel and Jay Prapaisilp noted the plaintiffs had done everything right, replying STOP and registering on the National Do Not Call Registry, and still lost that particular avenue. Their advice was not celebratory:

"Companies should not assume that other courts will adopt the Seventh Circuit's interpretation. Until further guidance emerges, continuing to treat text messages as subject to all applicable TCPA requirements is the safest bet."

Four weeks later came Zagury v. Puragin Water, 2026 WL 2299332 (S.D. Fla. Aug. 11, 2026). Eric Troutman of Troutman Amin put the question plainly:

"If a consumer responds 'Stop' to a text message does that mean you have to stop calling the consumer as well? According to one federal judge in Florida the answer may be 'yes.'"

The defendant conceded the STOP killed the texts and argued it left the calls alone. The court declined to draw that line at the dismissal stage, holding that the scope of a revocation depends on the surrounding circumstances rather than on which pipe it arrived through. As Receivables Info reported on August 17, the court reached for the FCC's cross-channel revocation requirement to get there.

Here is the part that should make any operator sit up. That FCC provision is not in effect. The agency waived it, and in January 2026 extended the waiver through January 31, 2027, while it decides whether to keep the rule at all. Troutman flagged the discrepancy with visible restraint:

"The Court noted that certain portions of the FCC's revocation rule have been delayed but, for some reason, concluded this portion of the rule was not (it was.)"

So a court applied a rule that does not apply yet, in a circuit whose neighbor just held the underlying channel is not even a call. And after McLaughlin v. McKesson, courts are no longer bound by FCC interpretation anyway, which is how you get two federal opinions in one summer that cannot both be operationalized by the same compliance memo.

Troutman's conclusion is the only sane synthesis, and it is worth reading twice because it dismisses the entire legal question as beside the point:

"Without question some courts will hold a stop request will apply across channels. Whether that conclusion is well founded is sort of irrelevant - if you want to avoid suits you may want to be conservative here."

The Real Problem Is Not Legal. It Is Architectural.

Every organization reading this has consent stored in at least three places. The CRM holds an opt-in checkbox from a web form. The dialer holds a DNC flag. The texting platform holds a STOP keyword suppression list. The email tool holds an unsubscribe. Four systems, four truths, zero arbitration.

That was survivable when a human was the one dialing, because the human remembered the angry text from Tuesday. It stops being survivable the moment you deploy an autonomous agent that reads a queue and acts.

BCG named this exact failure mode in a piece published August 14 on governing AI agents at scale. Writing for CIOs, Shaheer Rizvi and colleagues described what happens when agents proliferate faster than the rules around them:

"Agents quietly access systems they were never meant to reach. Sensitive data moves to external models without any safeguards. An agent built for one use case gets repurposed by another team, inheriting permissions that were never reviewed."

Their prescription is an Enterprise AI Control Plane: a governance layer sitting above every platform, providing common identity, a registry of every agent, and runtime policy enforcement that blocks non-compliant actions at the point of execution rather than in a post-mortem. The component worth stealing outright is what they call a golden path, a pre-governed template where identity, registration, monitoring, and policy enforcement come pre-wired, so the builder inherits compliance instead of implementing it. BCG's measured result: what took weeks of governed setup collapses to a day.

Real estate does not need an enterprise architecture team to apply that idea. It needs one file.

The Pattern Nobody Has Built: The Consent Ledger Your Agents Read Before They Speak

Here is the design. It is not a policy document and it is not a prompt. It is a single record that every agent must query before any outbound action, and which every agent must write to after any inbound one.

The principle: consent is not a property of a channel. It is a property of a person. Store it that way once, and the cross-channel question that just split two circuits becomes an implementation detail you already handled.

Layer 1. One record per human, not per channel

The load-bearing field is revoked_scope, and the load-bearing decision is its default. Set it to all_channels. The Florida court's reasoning turns on the idea that a consumer who says STOP probably meant stop, and the only clean way to rebut that is the mechanism the court itself described: a single confirmation message asking whether the request covers calls as well, with revocation_confirmed flipping to true only if the consumer answers. If they do not answer, the broad reading stands. That is not caution. That is the cheapest possible insurance policy, priced at one SMS segment.

Note also what is deliberately not a boolean. consent_written_proof is a URI because a list of phone numbers has never once won a TCPA case. The guidance ActiveProspect published on August 12 is blunt about the arithmetic: statutory damages run $500 to $1,500 per violation, and your defense is documentation of how, when, and where consent was captured.

Layer 2. The pre-flight check, expressed as an agent instruction

An Oppy is only as disciplined as the gate it has to pass through before it opens its mouth. This is that gate, written the way you would actually hand it to an agent:

Before any outbound message or call, query the consent ledger for this contact_id. Do not proceed if any of the following is true. First, revoked_at is not null and revoked_scope includes this channel. Second, revoked_at is not null, revoked_scope is narrower than all_channels, and revocation_confirmed is false. Third, the current local time at tz_resolved falls outside the permitted window for this contact's state. Fourth, contacts_rolling_24h is greater than or equal to 3. Fifth, consent_scope is informational_only and this message is marketing. If blocked, write the reason to the audit log and take no action. Never infer consent from the presence of a phone number in a CRM record.

Five conditions. Notice that not one of them asks the model to make a judgment call. This is the difference between a guardrail and a suggestion, and it is why the check belongs in the query layer rather than the system prompt. BCG's phrase for enforcement at the point of execution rather than through paperwork applies exactly here.

Layer 3. Inbound revocation detection that does not depend on a keyword

Carrier-level STOP handling catches STOP. It does not catch "please quit texting me," "wrong number," "take me off your list," or the same sentiment spoken aloud on a voice call at 8:47pm. An agent that reads and hears everything is the only listener that catches all of them, which makes agentic AI the fix for a problem that keyword automation created.

The instruction is short. Any inbound message or call transcript expressing a desire to stop being contacted writes revoked_at, sets revoked_scope to all_channels, sets revoked_via to sms_freeform or voice_verbal, and suppresses across every channel immediately. Not within ten business days. Immediately, because Florida's statute gives texters a fifteen-day cure window that only protects senders who can prove when they processed the opt-out, and a timestamp you can produce is worth more than a policy you can describe.

Layer 4. The state map, resolved at capture

Federal law is the floor, not the ceiling, and the floor is getting less relevant every quarter. The state statutes are where the exposure has migrated, in part because the Seventh Circuit's narrowing of federal text claims pushes plaintiffs' attorneys toward state causes of action. Pennsylvania's SB 992, signed July 12, brings the strictest quiet hours in the country and compliance is due around October 18, 2026.

If you cannot segment by state, the defensible national floor is 9:00a to 7:00p in the recipient's local time, weekdays only, three contacts per rolling twenty-four hours across all channels combined. That single configuration clears every row above. It also costs you the wider federal window in states that allow it, which is the trade every operator gets to price for themselves.

One more detail from that same guide, and it is the kind of thing that decides cases: resolve local time from something other than the area code. Number portability means a 212 number can be sitting in Los Angeles, three hours off your assumption, and plaintiffs' attorneys know it better than your dialer does.

Why This Is an Agent Job, Not a Policy Job

Jeremias Maneiro has run roughly 600 AI training sessions a year through his AI-Cademy program, and previously trained more than 7,000 agents at Douglas Elliman. Speaking to HousingWire's Jonathan Delozier on August 18, he identified the single biggest implementation failure in the industry, and it is not a technology gap:

"No question, training that stops at 'here's how it works' and never reaches 'here's the job you hand it.' Agents are impressed but they're not really equipped."

The job you hand it. That is the whole discipline. "Be compliant" is not a job. "Query the ledger, evaluate five conditions, write the audit row, and refuse to send if any condition fails" is a job, and it is a job a machine does better than a person at 8:47 on a Friday night. Maneiro also put a floor under the effort required, which deserves to be quoted at brokerage leadership everywhere:

"If you're going to spend six months and tens or hundreds of thousands of dollars on rolling out a new tech tool, you should spend double that, triple that on the training and implementation of it."

The leadership version of the same point came from HousingWire CEO Clayton Collins on the Power House podcast with Zeb Lowe on August 14, fresh out of the company's AI Summit. The episode's framing is that the question is no longer whether the technology works but what leaders do with it, and that executives are now managing token utilization and AI governance directly rather than delegating either downward. Consent architecture is precisely the sort of thing that gets built correctly when an executive owns it and never gets built at all when it is filed under someone's compliance checklist.

Which Oppy Does What

The ledger is one record, but the work splits cleanly across three agents, and the split is the point. Separation of duties is what makes an audit trail readable.

The Outreach Oppy is the one your revenue depends on, and it is the one with the least authority. It cannot write consent. It cannot clear a revocation. It can only ask permission and be told no. Anyone who has run a call center will recognize that as the correct org chart.

Everything above is buildable today on Oppy with phone, SMS, and email in one place and 70-plus tool integrations behind them, which matters here for an unglamorous reason: a consent ledger only works if it is the single place all channels check. Split the channels across four vendors and you have rebuilt the problem with better logos.

The Uncomfortable Advantage

Compliance work is usually framed as a tax. This one is not, for a reason that shows up in the same August reporting.

Maneiro described how consumers now choose professionals, and it is not with a phone book:

"In the future they're going to say, 'Here's the Realtors that I interviewed. Who should I hire?' If you're not part of that search, if you didn't come up in their ChatGPT search, and if your unique value isn't part of you using AI, then you won't be part of the selection that gets made, not necessarily by the client, but by AI."

An organization that can produce, on demand, a timestamped record of every contact attempt, every disclosure shown, and every opt-out honored within seconds is not just harder to sue. It is a materially better counterparty for a lender, a title underwriter, an insurance carrier, or a property management client running vendor diligence. The audit log is a sales asset with a legal side effect, which is a rare and pleasant shape for a project to have.

BCG's closing line about agent governance applies without modification to a brokerage, a mortgage shop, a title agency, a property management company, or the dental office down the street that just bought a texting platform:

"The window to do this right is narrowing."

Two courts spent this summer disagreeing about what five letters mean. The organizations that stop caring about the answer are the ones that already stored consent as a property of a person. Build the ledger. Hand the job to an agent. Let the circuits argue.

References

1. Seventh Circuit Creates New Uncertainty for TCPA Text Message Compliance. Shahin O. Rothermel and Jay Prapaisilp, Venable LLP, All About Advertising Law, July 20, 2026
2. Steidinger v. Blackstone Medical Services, No. 25-2398 (7th Cir. July 14, 2026). Opinion of the court, authored by Judge Kirsch
3. Court Holds A "Stop" Request to a Text Message May Revoke Consent For Both Calls and Texts. Eric Troutman, Troutman Amin LLP, August 13, 2026
4. Florida Court Says Text Message 'STOP' Request May Also Revoke Consent for Calls. Jager Robinson, Receivables Info, August 17, 2026
5. Enterprise AI Control Plane: The CIO's Guide to Governing and Accelerating AI Agents. Shaheer Rizvi, Zeeshan Shah, Sarvesh Singh, Nicolas Tanaka, Francois D'Agostini and Jazz Tobaccowalla, BCG, August 14, 2026
6. Why many real estate AI rollouts fail, and how to make them stick. Jonathan Delozier interviewing Jeremias Maneiro, HousingWire, August 18, 2026
7. Clayton Collins: What leaders are still getting wrong about AI. Power House podcast with Zeb Lowe, HousingWire, August 14, 2026
8. TCPA guidelines: Best practices for beginners. Marialuisa Aldeghi, ActiveProspect, August 12, 2026
9. State Texting and Telemarketing Laws Guide. Tychron, updated July 30, 2026
10. Pennsylvania SB 992: What Business Leaders Need to Know. TCPAWorld, July 14, 2026
11. FCC Delays Revoke All Consent Rule for Robocalls and Text Messages Until 2027. Venable LLP, January 14, 2026