A second signature is cheaper than a conservatorship

Your parent's voice is about thirty seconds of audio away from being cloned convincingly. Between doing nothing and taking legal control of their finances there is almost nothing on offer, and that missing middle is where most of the loss happens. The control that belongs there is small, boring, and entirely under your parent's command.

The call comes at 11 at night

The phone rings late, which is the first part of the design. A grandson's voice, upset and rushed, says there has been an accident, that he is in trouble, that he is embarrassed and please do not tell his mother. Another voice comes on, calm and official, and explains what needs to happen next and how quickly.

The voice is right. Not approximately right. Right. The catch in the throat, the way he starts sentences twice, the particular way he says his grandmother's name. Thirty seconds of a graduation video posted publicly two years ago is enough raw material for that now, and the tools that do it cost less than a takeaway meal.

She is not confused. She is not incompetent. She is a person who has just been told her grandson is in trouble, by her grandson, and every part of her wants to help him in the next ten minutes. She transfers the money herself, from her own bank, with her own credentials, correctly. Every control her bank has works exactly as designed, because there is nothing wrong with the transaction except its reason.

Her daughter finds out eleven days later. What she says next is the sentence that motivates this entire piece, and thousands of families say a version of it every week: I knew something like this would happen, and there was nothing I could have set up in advance to stop it.

How much money is this, and how much of it is now AI?

The scale is easy to underestimate because the losses arrive one household at a time. In 2025, people aged 60 and over in the United States filed more than 201,000 complaints and reported losing about $7.7 billion, a figure reported as up roughly 59 percent in a single year (FBI IC3 2025 Internet Crime Report). That is a larger reported loss than the entire business email compromise category in the same year.

The newer number is the one worth sitting with. The FBI's first year of tracking artificial intelligence enabled scams attributed about $352 million in losses among victims aged 60 and over to schemes involving synthetic voice and similar techniques, within a broader total of roughly $893 million in AI related losses across all ages (FBI elder fraud resources and the IC3 annual report). These are reported losses only. Under reporting in this category is severe, because the emotion that follows is shame.

Why the grandparent scam suddenly industrialised

This scam is decades old. What changed is the cost of the hardest part.

The old version required a caller who could improvise, sound young, and get lucky when the target was hard of hearing or half asleep. It failed most of the time, so it stayed a low volume crime. Voice synthesis removed the skill requirement. A short public clip from a graduation, a wedding toast, a podcast appearance or a social video is now sufficient to produce speech that sounds like a specific person, on demand, in real time, at essentially zero marginal cost.

When the cost of the hardest step in a crime falls to nearly nothing, the crime scales. This is the same shift that made voice fraud a top concern in commercial call centres, which we covered in the post on contact centre voice deepfakes. The difference is that a call centre has a fraud team, and your mother has a landline.

Why can the bank not simply refuse the transfer?

This is the question every family asks, usually with some heat, and the honest answer is more constrained than people expect.

A bank's default legal position is that a competent adult may move their own money. That is not a loophole. It is the foundation of the relationship, and the alternative, where institutions decide which of your instructions they believe in, has an uglier history than the fraud does. When a bank refuses a lawful instruction from a competent customer, it takes on real liability and real paternalism.

The financial industry has built genuine tools around the edges of that constraint, and it is worth knowing them precisely, because they are the foundation the missing control should be built on.

Under FINRA Rule 4512, broker dealers must make reasonable efforts to obtain the name and contact information of a trusted contact person for an account. Under FINRA Rule 2165, a firm that reasonably believes financial exploitation is occurring may place a temporary hold on a disbursement, and may contact that trusted person. The Senior Safe Act gives certain trained staff immunity for reporting suspected exploitation in good faith.

Read those two rules carefully and the gap becomes visible. The trusted contact person can be told. The firm can pause. Neither instrument allows the account holder to say, in advance and of their own free will, that certain transfers should require someone else's agreement. The industry built notification and discretion. It never built consent based co-authorization.

The three options families actually have today

What is available today is a cliff, not a slope.

Do nothing, and hope. This is what most families choose, because the alternatives feel like a betrayal.

Take partial or total control: a joint account, a power of attorney, in the worst case a guardianship or conservatorship. These work, and they cost something enormous. You are asking a competent adult to hand over a piece of their standing as an adult. Many parents refuse, and they are not being unreasonable when they do. Guardianship in particular has a documented history of abuse serious enough to have produced reform movements in multiple states.

Or add warnings and monitoring products: alerts to a family member, spending summaries, scam education. These help at the margins and share one flaw. They arrive after the transfer, or they ask a frightened person to doubt a voice they are certain they recognise.

What a signature can and cannot do here

This needs saying plainly and early, because the temptation in this category is to oversell, and families deserve better than that.

A signature cannot stop someone who genuinely intends to send the money. If your father is convinced, and he signs, the transfer is authorised and it should be. That covers a large share of investment fraud, romance fraud and long running confidence schemes, where the victim is not impersonated at all. They are persuaded, over weeks, and they act with full intent. No cryptographic control touches that, and anyone who tells you otherwise is selling something. Our analysis of which reported losses a signature would actually have prevented scores those categories honestly low.

What a signature does change is narrower and still significant.

It changes who must agree before money in a protected band moves. A cloned voice can convince your mother. It cannot produce a signature on your phone, three hundred miles away. That second requirement does two useful things: it inserts a person who is not currently inside the emotional pressure of the call, and it inserts a delay measured in minutes, which is a very long time in a scam engineered around urgency.

Almost every one of these scripts depends on isolation. Do not tell your mother. Do not discuss it with anyone. Stay on the line while you go to the bank. A rule that structurally requires a second person to be told is aimed precisely at that dependency, and it is aimed there by the account holder's own prior choice rather than by an institution's judgement.

What is a trusted co-signer rule?

Mechanically it is small, which is why it is disappointing that it does not exist yet.

The account holder sets a rule on their own account. Something like: transfers above $2,000, or any transfer to a payee I have never paid before, require my signature and my daughter's signature. The parent and the daughter each enrol a device once. The daughter never receives login credentials, never gets a card, and cannot see the balance or the account history. When a transfer matches the rule, both phones show the same short prompt: this amount, this payee, right now. Both people confirm with the gesture that unlocks their phone. The transfer proceeds.

Only the account holder can change or remove the rule. That single property is what separates this from every option in the lower half of the table, and it is not a detail. It is the whole ethical design.

What the co-signer actually sees

Minimal disclosure is a hard requirement, not a nice feature. The co-signer sees only what is needed to make one decision about one transfer:

{
  "action":       "transfer.approve",
  "account_ref":  "acct_7f21",          // opaque, not the account number
  "amount":       "4800.00",
  "currency":     "USD",
  "payee_name":   "SafeHold Recovery Services",
  "payee_ref":    "new_payee",          // never paid before
  "rule_matched": "new_payee_or_over_2000",
  "requested_at": "2026-09-04T23:11:52Z",
  "expires_at":   "2026-09-04T23:26:52Z",
  "nonce":        "b7c0a2e1"
}

No balance. No transaction history. No other payees. Just the one decision. The object is serialised canonically, hashed, and that hash is the challenge each device signs. Both parties see the same fields, and the signature covers those exact bytes, so nothing can be altered between what was shown and what was approved.

Notice payee_ref. The fact that this recipient has never been paid before is often the single most informative thing a family member could know, and it is exactly what nobody tells them today.

The rule check, at the transfer endpoint

async function approveTransfer(tx, holderRule) {
  if (!matchesProtectedBand(tx, holderRule)) {
    return proceed(tx);                     // ordinary spending, untouched
  }
  const payload = canonicalize(minimalDisclosure(tx));
  const [holderReceipt, cosignerReceipt] = await Promise.all([
    manav.requestSignature(holderRule.holderId,   payload),
    manav.requestSignature(holderRule.cosignerId, payload)
  ]);
  if (!holderReceipt.valid || !cosignerReceipt.valid) {
    return hold(tx, "co_signature_incomplete");
  }
  return proceed(tx, { receipts: [holderReceipt.id, cosignerReceipt.id] });
}

Three design points matter more than the code. The rule only touches the protected band, so groceries, utilities and existing payees are never affected. Both signatures cover the same payload, so the two people are provably agreeing to the same thing. And the receipts are Ed25519 signed objects that verify offline against a published key, which means that if this transfer is ever disputed, or ends up in an elder abuse investigation, there is durable evidence of exactly who approved what and when. The developer documentation covers the verification path, and the signing demonstration shows the flow.

Does this take away a parent's independence?

It is the right question to ask first, and the answer is genuinely no, for reasons that are structural rather than reassuring.

The parent sets the rule. The parent chooses the threshold. The parent chooses who the co-signer is. The parent can remove the rule at any time, alone, without the co-signer's agreement or knowledge. The co-signer cannot spend, cannot view, cannot move, and cannot lock anything. They can only decline to co-sign a transfer in a band the parent chose to protect.

The useful comparison is a seatbelt rather than a chauffeur. You still drive, wherever you want, whenever you want. You have simply decided in advance that one specific category of event should be harder.

There is also a quieter benefit that families report wanting: it gives an older adult a graceful exit line. "I would need to get my daughter to approve that, and she is asleep" ends a scam call, and it ends it without the person having to admit doubt or accuse the caller. Scripts of this kind are built to make refusal feel like an accusation. A rule makes refusal procedural.

What if the co-signer becomes the risk?

A significant share of elder financial exploitation is committed by family members, so a design that hands a relative new powers without examining this would be irresponsible.

The design answer is that the co-signer gains no power at all. They gain a veto over a band of transfers, and nothing else. There is no path from co-signer to spending, viewing, moving or restricting. The asymmetry is deliberate: the parent may remove the co-signer unilaterally at any moment, and the co-signer can never remove the parent.

The residual risk is coercion, where a relative pressures a parent into a rule and then refuses to sign, using the veto to obstruct. That is real, and it is why three properties are not optional: unilateral revocation by the holder, a visible record to the holder of every request and every refusal, and an institutional escalation path when a holder reports obstruction. Those are policy and product requirements, and any bank implementing this should treat them as such.

Honest limits

• It does not stop persuasion. Romance fraud, investment fraud and long confidence schemes usually involve a victim who fully intends the transfer. A co-signer who is also persuaded, or who trusts the parent's judgement, signs. This is a genuine and large share of the total loss.
• It requires institutions to build it. The rule lives at the bank or brokerage transfer endpoint. Today a family cannot switch this on unilaterally, which is why the practical action is to ask, repeatedly and in writing.
• It requires two people with devices. Enrolment is once, and can be done in a branch, but a household with no usable device for either party needs a different approach.
• It introduces a delay, and delay sometimes costs. A genuine urgent transfer at 3am with an unreachable co-signer will be held. Rules need sensible thresholds and a documented override that is slow enough to be safe.
• Cash and cheques route around it. A scam that sends someone to a branch counter, an ATM, a gift card rack or a cryptocurrency kiosk is a different control problem. Irreversible transfer paths deserve their own treatment, which we discuss in the post on authorising irreversible transfers.
• Policy engine features are not shipped. The signing, enrolment and receipts exist today. Rule evaluation currently lives in the relying party's own code, not in a packaged policy product.

What a family can do this week

1. Have the conversation about the rule, not about the risk. "I want to set up something on my account" is a very different conversation from "I am worried you will be scammed". Let the parent own it, because under this design they genuinely do.
2. Name a trusted contact person on every account. It is weaker than a co-signer rule but it exists today, it takes one phone call, and it is required to be offered on brokerage accounts under FINRA Rule 4512.
3. Ask each institution the specific question. Not "how do you protect seniors" but "can the account holder set a rule requiring a second person's approval for transfers over an amount they choose, revocable by the account holder alone". Ask in writing. Record the answer. Demand for a feature is measured in tickets.
4. Turn on every notification that goes to a second device. Alerts that reach only the account holder's phone are worth little if that phone is in the room during the call.
5. Agree a family verification word. Low technology and surprisingly effective against voice cloning, because the clone can reproduce a voice but not a shared secret it has never heard. Agree it out loud, in person, and never send it in a message.
6. Rehearse the exit line. "I never approve anything on a call. I will call you back on the number I already have." Said once in advance, it is much easier to say at 11 at night.
7. Know the reporting route before you need it. Report to ic3.gov and to the institution immediately, because the small chance of recall drops sharply within days.

If you build products at a bank or brokerage

The build is smaller than the compliance conversation around it. You need a rule object on the account, an enrolment flow for a second person who is not a customer, a payload with minimal disclosure, two signature requests, and storage of both receipts against the transaction. The regulatory story already exists in the shape of trusted contact and temporary hold obligations, and this is the consent based version of the same intent. The differentiator is that it is the customer's own instruction, which puts it on far firmer ground than a discretionary refusal.

Frequently asked questions

How can I protect an elderly parent from voice cloning scams without taking over their bank account? Ask their bank or brokerage for a co-signer rule, in which the account holder chooses a trusted relative and decides that transfers above a chosen amount, or to a payee they have never paid before, require both people to approve on their own devices. The parent keeps full control and can remove the rule alone at any time.

What is a trusted contact person, and is it enough? It is a person you name on an account whom the firm may contact if it suspects exploitation or cannot reach you, required to be requested on brokerage accounts under FINRA Rule 4512. It is worth having and it is not enough, because it only permits the firm to make a call at its own discretion. It gives your family no ability to require anything.

Can a bank stop my parent from sending money to a scammer? Usually not, and generally it should not be able to on its own judgement. A competent adult is entitled to move their own money, and a bank that refuses lawful instructions takes on liability and a troubling amount of power. Brokerages can place temporary holds under FINRA Rule 2165 when they suspect exploitation, but that is discretionary and after the fact.

How much audio does a voice clone need? Far less than most people assume. Public reporting and vendor demonstrations consistently describe convincing results from very short samples, on the order of tens of seconds, taken from ordinary sources such as social videos, voicemail greetings or recorded events. Treat any voice on an inbound call as unverified regardless of how familiar it sounds.

Does a co-signer see my parent's balance or transactions? No, and this is a hard design requirement rather than a setting. The co-signer sees only the single transfer they are being asked to approve: the amount, the payee name, whether it is a new payee, and the time. They receive no credentials, no card, no balance, no history, and no ability to move money themselves.

Sources

1. FBI Internet Crime Complaint Center, 2025 Internet Crime Report: more than 201,000 complaints and about $7.7 billion in reported losses among victims aged 60 and over, and the first year of reporting on artificial intelligence related scam losses.
2. FBI, Elder Fraud: description of grandparent and government impersonation schemes and reporting guidance.
3. FINRA Rule 4512: customer account information, including the trusted contact person requirement.
4. FINRA Rule 2165: financial exploitation of specified adults, and temporary holds on disbursements.
5. Consumer Financial Protection Bureau, resources for older adults: guidance and research on elder financial exploitation.
6. United States Department of Justice, National Elder Fraud Hotline: case management support for victims and families.

A cloned voice can convince your mother in ninety seconds. It cannot produce a signature on your phone, three hundred miles away.