.jpg)
For as long as elder law has existed as a practice area, verification of identity over the phone has rested on one simple, unspoken assumption: you know your own grandchild’s voice when you hear it. That assumption no longer holds. Artificial intelligence can now clone a familiar voice from as little as three seconds of audio, a birthday video, a voicemail greeting, a clip pulled from social media, or a robocall, and exploit it to power a version of the decades-old “grandparent scam” convincing enough to defeat even a cautious listener. The panic in the voice is real. The words are the right words. The voice itself is the only thing that isn’t.
This isn’t a hypothetical for the planning bar. It’s a documented, spreading pattern. Consumer-protection groups have spent recent weeks pushing families to adopt a “family code word” a phrase agreed on in advance, unrelated to anything posted online, to be used specifically when a call demands money or attention under pressure. It’s sound advice, and every elder law practitioner, financial planner, and insurance agent should be handing it out. But it’s advice aimed at the kitchen table. It doesn’t touch the legal architecture that actually controls whether money moves- the power of attorney, the trusted-contact designation, the account-hold authority sitting in a drawer, unconsidered- until the moment it’s tested.
That traditional planning architecture was built for a world where verification meant recognition. AI has broken that link, and it has broken it in two directions at once. A cloned voice can pressure the principal into authorizing a transfer. It can just as easily impersonate the agent, calling an institution directly and instructing it to move funds under an existing power of attorney the institution has no independent way to verify by voice alone. Most planning advice addresses only the first scenario. The second is at least as dangerous, because it bypasses the principal entirely and goes straight to the money.
States are beginning to notice the fallout, if not yet this particular version of it. Minnesota’s ban on cryptocurrency ATMs, which took effect August 1, was a direct response to roughly a million dollars in senior losses tied to these machines since 2023. Scammers often used the familiar emergency script to push victims into feeding cash into kiosks that converted it to untraceable crypto within minutes. The ban treats one symptom. It does nothing about the underlying vulnerability: once a caller sounds sufficiently convincing, most of our legal and financial safeguards still treat voice as authentication, whether that voice claims to be the principal in distress or the agent giving instructions.
Elder law planning has better tools available. They simply haven’t fully caught up to what AI has done to the threat model. Four drafting and advising changes are worth building into practice now, and it’s worth being explicit about which threat each one addresses.
Two-Party Authentication
First, treat unusual disbursements as a two-person decision, not a one-person judgment call, and write the requirement to bind the institution, not just the agent.
A standard power of attorney gives the named agent broad, immediate authority to act: the right design when the goal is avoiding paralysis during a medical crisis, and the wrong design when the emergency itself might be manufactured. For clients with meaningful assets, consider a threshold trigger: transfers above a defined dollar amount, or transfers requested under circumstances involving secrecy or urgency, require confirmation from a named second party (a co-agent, a designated confirmer, or a simple written acknowledgment) before the transfer is executed.
Communicate the restriction directly to the relevant financial institutions in advance, in writing, as a condition on the account rather than a private understanding between agent and principal. Drafted this way, the requirement protects against both threats: it slows down an agent who has been manipulated by a convincing call, and it stops an institution from honoring instructions from someone merely claiming to be the agent, since the institution itself is now contractually required to seek independent confirmation before acting.
Formalize Trusted Contact Designation
Second, formalize the trusted-contact designation at every financial institution a client uses.
FINRA Rule 2165 already gives broker-dealers the ability to place a temporary hold on a disbursement when financial exploitation is reasonably suspected, and to contact a client-designated trusted person before funds move. Many banks now offer comparable voluntary programs. The problem is that almost no one completes these forms until after something has gone wrong. Make this standard intake for every aging-in-place or elder law client: identify a trusted contact, confirm the designation is on file at each bank and brokerage, and revisit it the same way you revisit a beneficiary designation. Because the trusted-contact hold is triggered by the institution’s own suspicion rather than by who is on the phone, it functions as a backstop against agent impersonation as well as against pressure on the principal: the institution doesn’t need to know which version of the scam it’s looking at to use it.
Family Code Word
Third, put the “family code word” concept into the Trusted Contact Designation document itself, and extend it to cover instructions to institutions, not just calls to family.
A power of attorney or a supplemental letter of instruction can specify that no unusual or time-pressured transfer will be executed, whether requested by a purportedly distressed family member or communicated by phone to a bank or brokerage claiming to act under the power of attorney, without independent verification through a pre-agreed method: a callback to a known number, a code word, or confirmation through a second channel. Writing this into the governing document, and into the institution’s file on the account, does two things a verbal family agreement cannot: it gives everyone involved explicit cover to slow down and verify even under pressure, and it creates a standard a court or institution can point to later if a transaction is challenged as the product of fraud or undue influence, regardless of whether the fraud targeted the principal or impersonated the agent.
Settle a Trust
Fourth, consider a revocable living trust as a structural layer beyond the power of attorney, for reasons that go well past probate avoidance.
A funded trust changes who is actually handling disbursements and how they are made. A corporate or professional trustee typically already runs verification protocols scaled to resist exactly this kind of fraud as a matter of routine practice. Institutional trustees don’t disburse significant sums based on a single phone call from anyone, family member or agent, regardless of how convincing the voice. Even with an individual serving as trustee, a trust naming co-trustee, special trustees or a trust protector can require joint authorization for distributions above a set threshold, or from a specific account, building the same friction described above directly into the structure rather than relying on it being honored voluntarily.
Trusts also offer a privacy advantage that is easy to overlook and increasingly relevant to fraud prevention specifically. A will and recorded General Durable Power of Attorney both become a public record once it’s filed for probate or recorded, disclosing assets, beneficiaries, and family structure to anyone who looks. A revocable trust generally does not; its existence, its terms, its trustee and successor trustee names, and the value of what it holds are not filed anywhere as a matter of course. That matters because voice-cloning scams increasingly begin with reconnaissance, scraping social media, obituaries, and public records to build a convincing family narrative and identify who has assets worth targeting. A trust that never surfaces in a public filing gives that reconnaissance far less to work with.
One practical step follows directly from this: when funding a trust with real property, the deed conveying the property to the trustee is ordinarily recorded and, in most counties today, published in a searchable online index, which can reveal the trust’s name, the trustee, and by implication at least some details of a family’s private planning to anyone who searches the owner’s name. Where the local recorder’s office permits it, request that the deed be recorded without inclusion in the public-facing online index. Some counties offer this only case-by-case (for example for law enforcement or public officials) or only to certain categories of filer, so it’s worth confirming what a given recorder allows before assuming the option is available, but it costs nothing to ask, and it closes a gap that many practitioners don’t think to close.
Final Word
None of this requires new legislation and none of it depends on a state banning a particular payment method. It requires elder advisors and practitioners to recognize that AI voice cloning hasn’t just created a new scam; it has quietly invalidated an assumption baked into decades of standard drafting, and it has done so on both sides of the transaction. A person's "voice" used to be a reliable form of authentication, for the person asking for help and for the person authorized to give it. It no longer is either. Estate and elder law documents that still implicitly rely on “sounds like family” or “sounds like the agent” as a security check are already out of date, whether or not the families who signed them know it yet.
The clients who will be safest going forward aren’t the ones who happened to see a warning about grandparent scams on the news. They’re the ones whose planning was built by someone who understood that a convincing voice, on either end of the call, is no longer reliable proof of anything at all.
