This is the question every owner asks about ten minutes after they first hear their new AI receptionist answer a call, and the honest answer has two halves. The legal half is genuinely moving. The practical half has not changed in years: configure it to tell the truth, and there is nothing left to worry about.
The EU: Article 50, from 2 August 2026
The EU AI Act sets a transparency obligation for AI systems that interact directly with people. In substance, a person must be informed that they are dealing with an AI system, unless that is obvious to a reasonably well-informed and observant person given the circumstances. The date to know is 2 August 2026, when that obligation becomes applicable.
Two things about it are commonly misread. First, it is a transparency rule, not a ban: nothing in it stops an AI answering your phone. Second, “obvious from the context” is doing less work than people hope. A caller who dials a plumber and hears a fluent, natural voice pick up is not on notice of anything, and a synthetic voice that is good enough to be worth buying is precisely a voice that does not announce itself.
The United States: state by state
There is no federal AI disclosure statute for phone calls. California's SB 1001, in force since 2019, is the one most often cited: it prohibits using a bot to communicate with a person in California, with intent to mislead, in order to incentivize a purchase or influence a vote, without clearly disclosing that it is a bot. Several other states have since introduced AI disclosure requirements of their own, some naming a specific window at the start of a call.
The structural point matters more than any single statute: these rules generally attach to where the caller is. A business in one state taking calls from twenty is exposed to the strictest rule that applies to any of those callers, which is a good reason to pick a policy that satisfies all of them rather than to track them.
Recording is a different law, and it catches people out
Disclosing that the assistant is an AI does not cover recording the call. That is wiretapping law, and it is the one that produces actual claims. Most US states are one-party consent, but roughly a dozen require every party to consent; in the EU a recording is personal data and needs a lawful basis and a notice. If you switch recording on, the notice belongs in the first seconds of the call, separately from anything the assistant says about being an AI.
What this product does, by default
There are two settings and neither of them lies. The default, if asked, never volunteers it but answers straight the moment a caller asks whether they are talking to a person - one short sentence, in the caller's own language, and then straight back to helping them. The alternative, upfront, says so in the first breath after the greeting.
There used to be a third option that instructed the assistant to change the subject. We removed it. Under Article 50 it was the one configuration that could put an operator on the wrong side of the rule, and commercially it was worse than the problem it solved: a caller who suspects a bot and gets a dodge has been lied to by a system acting for your business, and being caught at that damages you far more than a synthetic voice ever will. For more on how callers actually react, see whether callers can tell it's an AI.
The practical answer
Set it to answer honestly when asked. Turn recording on only if you want it, and add the notice when you do. Keep the escalation path to a person short, because “can I speak to someone?” is the request that follows the disclosure question about a third of the time. Do those three things and the compliance question resolves itself in every jurisdiction we are aware of, without you tracking a single statute.
One caveat worth stating plainly: this page is a vendor's plain-language summary, current as of 11 August 2026, and not legal advice. Healthcare, financial services, debt collection and political campaigning each carry extra requirements this does not cover.