Restaurant AI Security — How Callra Handles Your Data

What security certifications does Callra AI hold?

Callra AI is pre-launch and early-stage, and we don’t hold formal third-party certifications yet — no SOC 2, no ISO 27001, no PCI DSS certification of our own, no HIPAA attestation. We’d rather tell you plainly what we do today than claim credentials we haven’t earned. Here is what is actually true right now.

As Callra completes formal audits, we will publish the real results here — not before.

How does Callra handle payment card data on phone orders?

When a guest reads a card number aloud during a phone order, the risk is obvious: a spoken card number could be recorded, transcribed, and stored. Callra’s approach is to keep raw card data out of our systems entirely. Payment is handled through hosted checkout and tokenization at PCI-compliant processors — Stripe, Converge/Elavon, Clover, or Square — and the card is tokenized before any Callra server processes it.

What Callra stores vs. what the payment processor stores

Callra stores a payment token, plus non-sensitive metadata — amount, last 4 digits, authorization time. Callra never stores the full card number. Refunds and reversals go through the processor’s API against the stored token, not through a card number Callra holds, because Callra holds none. Callra itself is not PCI DSS certified — the certified processors we route payments through are.

Handling of card-number audio

For the portion of a call where a card number is spoken, we aim to avoid retaining that audio segment or writing sensitive card data to transcripts. The stored record keeps only payment metadata — amount, token, last 4, and timestamp — not the card number itself.

Is Callra AI working toward SOC 2 or other certifications?

Formal certifications like SOC 2 Type II are audits of how a vendor’s controls operate over an observation period. Callra does not currently have an active SOC 2 engagement, and we won’t claim one until it’s real. As we grow, formal security audits are part of our roadmap, and we’ll publish verifiable results here — not marketing claims — once they exist.

How does Callra handle guest data privacy?

Guest data Callra processes — name, phone number, order history, voice recording, and transcript — is treated as data you own. Callra does not sell or share guest data with third parties. Requests about your data can be sent to privacy@callra.ai. Full detail lives in our privacy notice.

Call recording disclosure and consent

Calls may be recorded and transcribed for order accuracy. The AI agent can disclose that the call may be recorded and that the caller is speaking with an AI before capturing audio. Recording and consent behavior — including whether a disclosure plays before the first utterance — is configurable per state, so it can be set up to support one-party and two-party (all-party) consent requirements. This is a product feature and a compliance commitment we are building toward, not a completed legal certification.

An example disclosure

State-by-state two-party consent landscape

Several states require all-party (two-party) consent before a call can be recorded. Callra’s disclosure and recording settings are configurable so an operator can align call handling with the requirements of the state they operate in.

This information is provided for general awareness and is not legal advice. Operators should confirm recording-consent requirements for their state with their own counsel.

Data residency — where is restaurant call data stored?

Callra hosts data with AWS in the United States. Voice recordings, transcripts, and order data are encrypted in transit and at rest.

Voice AI architecture — how Callra separates payment data

The reason Callra can keep raw card data out of its own systems is architectural: the system is built as separated layers, so the components that handle payment never share state with the components that record and transcribe audio. The diagram below shows the pipeline and the point where card data leaves Callra’s systems for the payment processor.

Layer separation — model, orchestration, tools, and logging

The model layer, the orchestration layer, the tools layer, and the logging layer are kept separate so that logging never sees a full card number, and the tokenization boundary sits at the payment processor. A guest call enters over telephony infrastructure, the speech-recognition layer interprets the order, the menu-lookup layer reads from the POS source of truth, and payment is tokenized at the processor boundary. Only a token and the last 4 digits are written back to the POS.

Security testing

Callra does not currently run a scheduled third-party penetration-testing program. Security review and testing cadence are part of our roadmap as the product and company mature.

Accessibility

We aim to make Callra’s web experience accessible, and the voice AI supports escalation to a human or text channel. We do not currently hold formal WCAG certification and won’t claim one until an audit confirms it.

Which subprocessors handle call data?

Callra uses a focused set of subprocessors to run the product. Here is the full list of providers that touch call audio, transcripts, or customer data:

Questions about our security practices?

If you have questions about how Callra handles your data, email privacy@callra.ai, or book a demo and our team can walk you through our current practices and our roadmap for formal certifications.