AI Dispatch Compliance: Calls, Consent, and Records

When an AI system places dispatch calls, sends texts to drivers and brokers, or records conversations to build a paper trail, the carrier is the one on the hook for how those communications happen. Federal and state rules on call recording, automated outreach to cell phones, and opt-outs were written for humans, but they apply just as much when a model is doing the dialing. None of this is a reason to avoid AI in dispatch. It is a reason to set up the consent, opt-out, and recordkeeping plumbing once, correctly, so the speed you gain does not turn into a liability you did not see coming.

This is a practical map of the areas to get right, not legal advice. Rules differ by state, change over time, and turn on facts specific to your operation. Treat everything below as a checklist to take to your own counsel, not a substitute for them.

Call recording consent is a state-by-state problem, not a single rule

The first thing to understand is that there is no single national standard for recording a phone call. Federal law and a majority of states follow a "one-party" rule, meaning one participant on the call can consent to recording it, and if your dispatcher or your AI is on the line, that participant can be you. But a meaningful set of states require "all-party" consent, where every person on the call has to agree before recording is lawful. The hard part for a carrier running interstate lanes is that you are constantly placing calls across state lines, and the safe assumption is that the stricter rule can apply when any party to the call is sitting in an all-party state.

For an AI that records by default to create a transcript or a quality log, this matters a lot. A human dispatcher in a one-party state who occasionally records is a narrow exposure. A system that records every call automatically, to brokers and drivers who may be anywhere in the country, is a standing one. The practical move most teams land on is to record with a clear disclosure on every call regardless of where it originates, because a disclosure plus the other party continuing the conversation is the cleanest way to cover both one-party and all-party situations. That is also why you hear "this call may be recorded" on so many business lines: it is the lowest-friction way to stop having to know each caller's state.

There are second-order questions worth raising with counsel before you flip recording on across the board. Does your disclosure need to be at the very start of the call, before anything substantive is said? Does an AI-generated voice change what you have to disclose, given newer rules treating artificial voice as its own category? Who owns and can access the recordings internally, and for how long are they kept? Getting a one-line answer to "should we record everything, and how do we disclose it" is the single highest-leverage compliance decision in this whole area, and it is worth a short, specific conversation with a lawyer rather than a default setting you never revisit.

Automated calls and texts to cell phones are their own regulated category

Recording consent is about capturing a conversation. A separate body of federal rules governs whether you can place the automated call or send the text at all when the number is a wireless one. The shorthand people use is TCPA, and the part that bites freight operations is the treatment of automated calls and texts to cell phones, especially when there is any prerecorded or artificial voice involved or when messages go out at volume. The point is not to memorize the statute. The point is that "we'll have the bot blast our driver list" and "we'll auto-dial every broker who posted a load" are exactly the patterns these rules were built to constrain.

The distinction that does most of the work in practice is the relationship between you and the person you are contacting. Reaching your own contracted drivers about a load they are already on sits in very different territory than cold automated outreach to phone numbers you scraped or bought. Express consent, the nature of the prior relationship, and whether the message is informational or promotional all change the analysis. A model that does not know the difference between "text my driver his next stop" and "text two hundred carriers I have never worked with" is a model that can walk you straight into the riskiest version of this. So the design question is whether your system can tell those cases apart and apply different rules to each.

This is the area where the cost of getting it wrong is concentrated, because automated-contact rules carry real federal exposure and attract litigation. I am deliberately not quoting penalty figures, because the numbers move and the math depends on facts like how many messages went out and whether violations are treated as willful. What you actually need from counsel is narrower and more useful: a clear line on which of your automated-contact flows are fine as-is, which need express consent captured first, and which you should not automate at all. Bring them the specific flows, not the abstract question.

Honoring opt-outs has to be a system property, not a manual step

Whatever the rules permit you to send, the other side keeps the right to make you stop. A driver who replies STOP, a broker who says "take me off your list," a contact who asks not to be called again, each of those is an opt-out you are expected to honor, and to honor reasonably quickly. When the outreach is automated, the opt-out handling has to be automated too, because a stop request that lands in a transcript nobody reads is the same as no opt-out at all.

The failure mode to design against is the gap between channels. Someone texts STOP and the texts stop, but the AI keeps calling them. Or a broker tells your dispatcher on a recorded call to quit emailing, and that instruction never propagates to the automated email flow. Opt-outs need to be captured wherever they arrive, including in the body of a conversation rather than as a tidy keyword, and then suppressed everywhere your system might reach that person. That is a real engineering requirement: a shared suppression list that every outbound channel checks before it sends, and a way to catch opt-out intent expressed in plain language, not just the literal word STOP.

It is also worth keeping a record of the opt-out itself, when it came in and when you honored it, because if a dispute ever arises the question will be whether you stopped and how fast. This is one of the clearer places to keep a human in the loop on edge cases: an ambiguous "stop sending me the bad loads" is not necessarily a blanket opt-out, and a person should make that call rather than a classifier guessing. Build the system to honor the obvious cases instantly and to flag the ambiguous ones for review.

Keep the records that prove what was agreed and what happened

Compliance is not only about restraint on the outbound side. It is also about retaining the right evidence so that, months later, you can reconstruct what was agreed and by whom. Freight is full of disputes that come down to "what did the rate con say" and "who approved that." When an AI is negotiating and booking, the bar goes up, because you want a clear trail showing a human authorized the commitments that mattered and showing the terms each party actually agreed to.

A workable retention list is short and concrete. The point is to keep enough to settle a dispute and to show your consent and opt-out practices were followed, without hoarding so much sensitive data that storage itself becomes a liability.

That last row is worth dwelling on, because the records you keep do double duty as a fraud defense. With CargoNet reporting cargo theft losses of $725 million in 2025, up roughly 60 percent year over year across 2,646 incidents, and double-brokering on the rise, the broker and carrier identity trail you retain is part of how you catch a bad actor before the load moves. Retention here is not bureaucratic overhead. It is the same evidence that protects you in a payment dispute and in a theft investigation.

How long to hold each category, where to store it, and how to secure personal information in driver and broker records are questions to settle with counsel and to write down as a policy, not to decide ad hoc per load. The federal posture on consumer data is that companies should keep the privacy promises they make and secure data in proportion to its sensitivity, which is a reasonable internal standard to hold yourself to even where a specific rule does not force it.

Where the human stays in control, and where the model genuinely helps

The honest version of this is that AI does not remove the carrier's legal responsibility for these communications; it concentrates it. Every call placed, text sent, and recording made at machine speed is still your call, your text, your recording. So the parts that carry legal weight, deciding to record across all states, defining which automated-contact flows are permitted, ruling on an ambiguous opt-out, authorizing a booking commitment, are exactly the parts to keep a person on. A model should surface these decisions, not quietly make them.

Where the system earns its place is in making the compliant path the easy one. It can attach the recording disclosure to every call automatically, so no dispatcher forgets. It can check a shared suppression list before any channel sends, so a STOP in one place stops everything. It can flag ambiguous opt-outs for a human instead of guessing, and it can build the retention trail, the rate cons, the logs, the approvals, as a byproduct of doing the work rather than as a chore someone has to remember. Done this way, automation makes you more compliant than a purely manual operation, not less, because consistency is exactly where humans slip and software does not.

The takeaway: get four things settled with your own counsel before you let AI touch dispatch communications at scale, namely your recording-and-disclosure rule across states, which automated-contact flows are allowed, how opt-outs are captured and honored everywhere, and what you retain and for how long. Then build those answers into the system so the rules are enforced by default rather than left to memory. See how AI Hub keeps a dispatcher in control of negotiation and booking while the routine work runs underneath.