Voice commerce is taking off, with bots from everyone from Meta to SpaceX seeking to add voice to how people sell. But with that comes a big problem, warns Ron Kerbs, founder and CEO of Kidas
TL;DR
- Voice commerce needs trust to grow: AI agents and cloned voices make it harder to know who—or what—is calling, creating opportunities for fraud and wasting businesses’ time.
- Verified caller ID is only part of the answer: STIR/SHAKEN authenticates the calling number, but even properly signed calls can carry scams.
- Protection must continue beyond onboarding: Kerbs argues that built-in synthetic voice detection and real-time traffic monitoring are essential to identify suspicious activity and protect legitimate voice commerce.
For most of the telephone’s history, hearing a voice meant something. You could pick up on hesitation, warmth, a familiar accent. A voice was proof that a real person was on the other end of the line.
That assumption no longer holds. A convincing voice can now be generated in minutes and the people who exploit trust for a living have noticed. Regulators have too. On July 22, the Federal Communications Commission (FCC) in the US proposed a major expansion of its Robocall Mitigation Database, including stricter screening of providers before their filings take effect and possible financial guarantees meant to keep shell companies and repeat offenders off the network.
The message for everyone in the voice value chain is clear. Responsibility for what travels across the network is growing, and the threat is changing faster than the rules.
What is voice commerce?
Voice commerce is any transaction that happens through a voice interaction, such as buying, paying, booking, or proving who you are. For decades, that meant call centers and interactive voice response (IVR) payment lines. Today it increasingly involves AI agents.
In late September, Meta announced real-time voice conversations for its Muse assistant, an email address for each user’s agent and retail partnerships with Walmart, Best Buy, Gap, Sephora, and Wayfair. The same week, SpaceXAI announced that Grok Bot, the assistant built into X, is getting voice calls and the ability to work with credentials stored in a user’s 1Password vault.
Put those capabilities together and you get a new kind of caller: an AI agent with a voice, access to your credentials and connections to retailers. Soon a meaningful share of calls will have a machine on one end, and eventually on both.
Not every one of those calls is a threat, and many will be legitimate agents doing legitimate work. But customers already tell us their sales teams lose real time and money talking to inbound AI agents without realizing it. Before we even get to fraud, the industry has a more basic problem: knowing who, or what, is on the line.
What voice fraud is and why it’s rising
Voice fraud is any scheme that uses a phone call to deceive. It includes impersonation of banks, government agencies and family members in distress; cloned and deepfake voices; spoofed caller ID; social engineering aimed at call center staff; and AI powered robocalls running at a scale no human operation could match.
Several forces are driving it up. The tools to clone a voice or run an automated calling campaign are cheap and widely available. More importantly, people have spent their whole lives learning to trust a voice and to assume it belongs to a real person, which is why deceiving someone by voice is still easier than text. Because a call happens in real time and carries emotion, victims often decide before they even have a chance to think.
Voice fraud is also difficult to fight, because it often leaves no trace. A phishing text or an email sits in an inbox as evidence, and most calls are never recorded. When a victim reports a scam call, there is often nothing to show what was said, which makes these crimes harder to prove, harder to report and harder to prosecute.
The industry’s main defense against caller ID spoofing is STIR/SHAKEN, a set of standards for signing and checking calls. When a call starts, the originating provider attaches a digital signature stating how well it knows the caller. Full attestation means it knows the customer and their right to use the number. Partial means it knows the customer but not the number. Gateway means it is only passing the call along. The terminating provider checks that signature before the call reaches the recipient, which makes spoofed numbers easier to catch.
But STIR/SHAKEN only answers one question: can this caller ID be trusted? It says nothing about what the call is trying to do. A properly signed call can still be a scam, and we see this happen in two ways:
- Legitimate companies are hacked, and their trusted traffic is used to place fraudulent calls.
- Bad actors, often behind shell companies, mislead providers into issuing them valid credentials.
This is not theoretical. In August 2024, the FCC reached a $1 million settlement with Lingo Telecom. The company had given an “A” attestation, the highest level of trust, to spoofed robocalls that used an AI-generated clone of President Biden’s voice to tell New Hampshire voters not to vote in the January 2024 primary.
What the industry needs to do
No single fix will stop voice fraud, but two changes would make the biggest difference. First, detecting AI-generated voices should be built into every consumer and enterprise voice service. Email providers don’t ask users to opt in to spam filtering, because it’s simply part of the service, and voice deserves the same treatment.
Phones, contact centers, bank lines and voice commerce platforms should all be able to tell whether the voice on a call is synthetic. The point is not to block every AI call. It’s to give people and businesses what they need to decide whether to trust it.
Second, voice providers need to monitor traffic in real time instead of relying only on know-your-customer (KYC) checks. KYC is a gate a customer passes once, at onboarding. It cannot catch an account that gets compromised next month, a shell company that cleared screening, or a bad actor holding valid credentials. What a customer’s calls actually do, day after day, is a far more reliable signal. Providers that analyze traffic as it flows can stop scam calls before they reach a victim, rather than learning about them from a regulator.
The FCC’s latest proposal raises the bar at the front door, which is welcome, but fraud that begins after onboarding will only be caught by watching what moves through the network.
Trust is the whole business
We have analysed millions of conversations, and the lesson is consistent: the clearest warning sign shows up in what a call does, not in who appears to be calling. Voice commerce has enormous potential. It can make buying, booking and getting help as easy as having a conversation, but it will only scale if people can trust the voice on the other end. In a world of AI agents, that means knowing what a call is doing, not just who signed it. The whole value chain owns that responsibility.
Author
Ron Kerbs is founder and CEO of Kidas
