PCI DSS
Not legal advice. This page describes legal and regulatory frameworks in general terms. Jurisdiction changes the answer, and the same voice deployment can be lawful in one market and unlawful in another. Take advice on your specific circumstances.
In one sentence
PCI DSS is the contractual security standard the payment card industry imposes on anyone who stores, processes or transmits cardholder data, and for a voice agent the compliant approach is almost always to make sure it never touches card details at all.
Not to be confused with HIPAA, or SOC 2.
Definition
PCI DSS is the rulebook the card networks impose on anyone who stores, processes or transmits card data.
If a voice agent handles card details, the standard governs how. The simplest way to comply is to build so the agent never hears them.
The Payment Card Industry Data Security Standard is a contractual framework set by the card networks rather than a law passed by a legislature. It applies to any entity that stores, processes or transmits cardholder data, and voice channels carry their own specific and long-established requirements.
The core prohibition that shapes everything else
- Sensitive authentication data, the card verification value among it, must not be stored after authorization. Not in a recording, not in a transcript, not in a log.
- That single requirement is what makes naive call recording incompatible with taking card payments, because a recording of somebody reading out a card verification value is stored sensitive authentication data.
The established mechanisms for handling it
- DTMF capture with suppression. The caller enters the digits on the keypad, the tones are stripped from the recording, and the digits pass directly to the payment processor. Neither the agent nor the recording ever holds the card data.
- Pause and resume recording. The recording stops for the card capture. Simpler, and more error-prone, because a pause that fails to fire stores card data.
- Redirect to a separate payment channel, for instance a secure link or a callback to an automated payment line.
- Descoping is the objective to design toward. The fewer parts of the environment that ever touch cardholder data, the lighter the assessment burden.
What voice AI adds to the problem
- Recognizing a spoken card number puts the digits through the speech recognition provider, the transcript store and potentially the model provider. Each one enters the cardholder data environment, which is an enormous expansion of scope.
- That is a strong practical argument for keypad entry or a redirect, and it holds entirely independently of the separate accuracy argument for keypad entry.
- On a website widget there is no keypad, so the equivalents are a text entry field or a handoff to a payment page. A voice agent that accepts spoken card numbers on a website has made a significant compliance choice, most likely without meaning to.
Scope and validation
- Validation requirements vary with transaction volume and merchant level, with self-assessment questionnaires for smaller merchants and formal assessment for larger ones.
- Which questionnaire applies depends on how payments are handled, which is why descoping changes the compliance burden rather than merely the risk.
Common misconception
That PCI DSS is a law with regulators behind it. It is a contractual standard enforced through acquirer and card network relationships. The consequences of failing it are commercial and contractual: fines passed down through acquirers and the possible loss of card processing, which in practice arrives faster than most regulatory enforcement.
Why it matters commercially
Any voice agent in a commerce setting will eventually be asked to take a payment. The correct architectural answer was settled long before AI voice existed: the agent should not hold card data at all. Designing for that from day one heads off a scope problem that is painful to unwind later.
In voice specifically
Speech creates the exposure by working correctly. Recognition has to succeed for the agent to be useful, and every success writes the card number into a transcript and through every provider that produced it. The keypad route contact centers rely on does not exist in a browser, so the equivalent has to be a text field or a handoff to a payment page.
Where AsqVox fits
The Orb does not process payments, so none of this applies to its live feature set today. It is covered here because descoping is the right default for any voice deployment, and because business owners ask about card handling early.
Visual
The card data your agent should never hear
The goal is not to secure the card data. It is to never have it.
Statistics
Every figure carries its source and year. Vendor numbers are labelled as vendor numbers, and where no reliable figure exists this page says so rather than borrowing one.
PCI DSS is a contractual standard maintained by the PCI Security Standards Council and enforced through card network and acquirer relationships, not a statute.
Contractual, not statutoryindustry rangePCI Security Standards Council, 2026 - The distinction changes who enforces it and how quickly. Fines pass down through acquirers and card processing can be withdrawn, which lands sooner than most regulatory action would.
Sensitive authentication data, including the card verification value, must not be retained after authorization.
No storage after authorizationindustry rangePCI DSS, 2026 - This one requirement is the reason naive call recording and card payment handling cannot coexist. It applies to a recording, a transcript and a log alike.
DTMF suppression and pause-and-resume recording are established mechanisms for keeping card data out of call recordings.
Two established mechanismsindustry rangePCI DSS and contact center practice, 2026 - Pause and resume is the weaker of the two, because a pause that fails to fire stores card data and nobody finds out until the recording is reviewed.
Validation requirements vary by merchant level and by how payments are handled, with self-assessment questionnaires applying to smaller merchants.
Varies by merchant levelindustry rangePCI DSS validation requirements, 2026 - Which questionnaire applies is a function of the architecture, which is why descoping reduces the assessment burden and not only the exposure.
Standard versions and requirement numbering change between releases, so any page citing specific requirement numbers goes stale.
-no reliable figureThis is why no clause number appears anywhere in this entry. Link to the current standard rather than restating numbers that move.
No reliable published statistic exists for PCI-related incidents involving AI voice agents specifically.
-no reliable figureThe category is young and incident reporting is not broken out this way. Any figure you are shown is vendor telemetry, and should be labeled as such.
Examples
In practice
A commerce voice agent is designed to take card details by speech, on the argument that recognition accuracy is now good enough. Security review stops it. The transcript store, the speech recognition provider and the model provider would all enter the cardholder data environment, expanding assessment scope enormously, and transcripts would hold stored sensitive authentication data. The design changes to a handoff to a payment page and the scope problem disappears entirely.
The everyday version
PCI DSS is the rulebook for taking card payments. For a talking agent the key point is simple: it should never hear the card number in the first place. A recording of someone reciting their security code is precisely what the rules forbid. Send them to a payment page, or let them type it in.
Usage
Who says it
- Payment and security compliance teams, who use it precisely.
- Contact center operations, where DTMF suppression is long-established practice.
- Commerce and finance procurement.
Where it turns up
- Usually in the same clause as payment handling, DTMF suppression support, recording controls, cardholder data environment scope and attestation of compliance.
- In regulated commerce, a vendor unable to demonstrate descoping is often excluded outright.
Common misuse
- Treating PCI DSS as a law. It is contractual, which does not make it any less binding.
- Assuming better speech recognition makes spoken card capture sensible. Accuracy was never the primary objection.
- Relying on pause and resume without failure handling. A missed pause stores card data.
Questions people ask
Is PCI DSS a law?
No. It is a contractual standard maintained by the PCI Security Standards Council and enforced through card network and acquirer relationships. That does not make it less binding. The consequences of failing it are commercial: fines passed down through acquirers and the possible loss of card processing, which usually arrives faster than regulatory enforcement would.
Can a voice agent take card payments?
It can, but every compliant design keeps the card data away from the agent. DTMF capture with suppression sends typed digits straight to the payment processor with the tones stripped from the recording. Pause and resume stops the recording during capture and is more error-prone. A redirect to a payment page or an automated payment line is the simplest descoping. Spoken capture pulls the recognition provider, the transcript store and the model provider into the cardholder data environment.
Why can a call recording not contain a card security code?
Because sensitive authentication data, including the card verification value, must not be retained after authorization. A recording of somebody reading out their security code is stored sensitive authentication data, and the same applies to a transcript or a log. This is the requirement that makes naive call recording incompatible with taking card payments.
How does PCI descoping work on a website widget with no keypad?
The keypad route is not available in a browser, so the equivalents are a text entry field or a handoff to a payment page. A voice agent that accepts spoken card numbers on a website has made a significant compliance choice, usually without intending to, because the transcript, the recognition provider and the model provider all enter the cardholder data environment the moment it does.
Last reviewed 30 July 2026. Written and reviewed by Dhruv Dholakia, founder of AsqVox.