Identity
The Voice Passport
Every AI agent carries a signed, portable credential — issuer, model, purpose, and policy — checked against the registry at the start of every call. Spoofed agents never connect.
The trust layer for AI voice
Before AI speaks, someone has to verify it.
Millions of phone calls are now made by machines. NoAICalls sits between every AI agent and every ear — verifying identity, disclosing AI, capturing consent, and sealing an audit trail before a single word is spoken.
- Agent → Layerpresent voice passport vp_9f3a24c1sig ed25519
- Layerverify signature against passport registry✓ 118 ms
- Layer → Caller“This call uses an AI agent. Recording requires your consent.”disclosed
- Caller → Layerconsent granted — recording✓ logged
- Sessiontrusted session established · audit log_7c21 open · human escalation armed
Verification in <200 ms SOC 2 program underway Open protocol specification Maps to EU AI Act Art. 50
01 The shift
The next billion phone calls won't come from people.
Support lines, appointment desks, collections, sales, dispatch — AI agents are already dialing and answering at scale. The technology arrived fast. The trust to go with it didn't.
24/7autonomous agents never hang up
~0¢marginal cost of a cloned voice
1question every caller now asks
02 The problem
And the person answering can't tell.
Every unanswered question above is a hang-up, a complaint, or a regulator's exhibit. Today's voice infrastructure answers none of them. It connects audio. It doesn't establish trust.
AI agent · 00:04“Hi, I'm calling about your account ending in 4417…”
Caller · 00:09“Wait — is this actually my bank?”
Caller · 00:12“Am I talking to a machine?”
Caller · 00:15“Is this being recorded? Can I get a human?”
— silence. the system has no answer. —
03 The cost of no trust
Unverified voice is already expensive.
Fraud
Cloned voices made deception free.
In 2024, deepfaked executives on a single call moved US$25M out of one multinational's Hong Kong office. Without verifiable identity, every call is spoofable — and callers know it.
Compliance
The rules aren't coming. They're in force.
The FCC ruled AI voices count as “artificial” under US robocall law in 2024. The EU AI Act's duty to disclose AI interaction — Article 50 — applies from 2 August 2026, with fines up to €15M or 3% of global turnover.
Trust
Doubt ends conversations.
When a caller can't tell what they're talking to, they hang up, escalate, or churn. Adoption stalls not because AI fails — but because trust does.
The forcing function · regulatory timeline
2019 · CaliforniaFirst bot-disclosure law: automated agents must identify themselves.
Feb 2024 · United StatesFCC: AI-generated voices are “artificial” under robocall law — consent required.
2024–25 · US statesUtah and a widening set of states mandate AI disclosure to consumers.
2 Aug 2026 · European UnionAI Act Art. 50 applies: people must be told they're talking to AI. In force next month.
Penalty tier for Art. 50 breaches: up to €15M or 3% of worldwide turnover. NoAICalls turns the mandate into an API call.
04 The missing layer
The web got a padlock. The phone never did.
In 1994, SSL gave the web verifiable identity and encryption — and the padlock made online banking, shopping, and medicine possible. Voice is having its 1994 moment. NoAICalls is the padlock — plus the paperwork behind it.
05 How it works
One layer between every AI and every ear.
NoAICalls deploys inline. Conversations don't route around it — identity, disclosure, and consent are established before the customer hears a word.
Enterpriseyour agents & policies
Voice platformsvapi · retell · custom
NoAICallsverify · disclose · consent · audit
Telecomsip · pstn · carriers
Customera call worth trusting
Inline by design — zero conversations bypass verification.
06 Platform
Everything a trusted call requires.
Consent & disclosure
The Conversation Ledger
AI disclosure is announced, consent is captured in the caller's own words, and every event is written to an immutable, exportable ledger — timestamped, sealed, audit-ready.
Human escalation
A person, one request away.
“Can I talk to a human?” is honored instantly — by voice or keypress — with full context handed to the agent. Escalation isn't an apology. It's part of the protocol.
VT-04Trust CertificatesEarned verification badges that prove your AI program meets the standard — presented at call time.
VT-05Compliance HubMap controls to frameworks, monitor drift, export evidence on demand.
VT-06Trust ScoreA live measure of transparency and conversation quality across your fleet.
VT-07Fraud DetectionFlag spoofed passports, rogue agents, and anomalous behavior in real time.
VT-08Reputation EngineLongitudinal trust analytics for every agent, campaign, and line of business.
07 Who it's for
Built for the people on the hook.
Four owners, one layer. Each gets an answer — and a number that proves it.
Compliance · Legal
“Can we evidence every disclosure and consent — per call, per regulator?”
Exportable, hash-sealed records mapped to TCPA and EU AI Act Art. 50. Evidence on demand, not on request.
MEASURED BY · consent coverage % · audit export time
Risk · Security
“How do we stop a cloned voice speaking as us?”
Signed passports checked against the registry on every call. Spoofed and rogue agents never connect.
MEASURED BY · % calls verified · spoof attempts blocked
CX · Operations
“Will customers actually stay on the line with AI?”
Honest disclosure plus a guaranteed human path turns suspicion into tolerance — and tolerance into resolution.
MEASURED BY · hang-up rate · escalation SLA
Engineering · Platform
“What does this cost us to run?”
Three endpoints, inline, under 200 ms added latency. No re-platforming, no lock-in on your voice stack.
MEASURED BY · added latency · days to integrate
08 Developers
Ship trust in an afternoon.
Wrap your existing call flow — no rip-and-replace. Three endpoints cover the whole handshake.
POST /v1/verify
POST /v1/consent
GET /v1/audit/:id
- SDKs for Node, Python, and Go — typed, retried, boring in the good way.
- Webhooks for consent, escalation, and trust-score events.
- Sandbox passports and a full mock registry from day one.
verify.ts
import { NoAICalls } from "@noaicalls/sdk"; const nac = new NoAICalls(process.env.NAC_KEY); // the whole handshake, one call const session = await nac.calls.verify({ agentId: "agent_meridian_care_01", disclose: true, consent: ["recording"], escalation: "on_request", }); session.trust; // { verified: true, latencyMs: 118, // passport: "vp_9f3a24c1", // audit: "log_7c21" }
09 Security & compliance
Engineered like the infrastructure it is.
A trust layer has to be more trustworthy than what it protects. Security isn't a page on this site — it's the reason the product exists.
Encryption everywhereTLS in transit, AES-256 at rest, keys rotated.
SOC 2 programType II underway; report available under NDA.
Immutable audit logsHash-chained, timestamped, export on demand.
RBAC & SSORole-based access, SAML/OIDC, SCIM provisioning.
Consent recordsQueryable, revocable, retained to your policy.
Anomaly alertsReal-time flags on spoofing and policy drift.
10 Due diligence
Ask the hard questions.
We'd rather answer them on the homepage than dodge them in a deck.
Q·01Can't the voice platforms build this themselves?
Features, yes. Trust, no. Verification issued by the party being verified isn't trust — it's marketing. Like certificate authorities on the web, a trust layer only works because it sits outside the systems it vouches for — and works across every platform and carrier at once. Independence is the product.
Q·02Is NoAICalls a product or a protocol?
Both, deliberately. The Voice Trust Protocol is open — anyone can implement it, anyone can verify it. The registry, enforcement, and enterprise tooling around it are the business. SSL worked exactly this way: open standard, commercial infrastructure on top.
Q·03What if the regulation slips?
Fraud doesn't wait for effective dates, and neither do customers. Cloned-voice scams and silent hang-ups are on this quarter's P&L, not 2027's. Regulation sets the floor. Trust sets the win rate.
11 Vision
Trust is the protocol.
SSL didn't win because it was clever. It won because it was shared. That's why we're publishing the Voice Trust Protocol (VTP) as an open specification — a trust layer only works if anyone can verify it. Voice first. Messaging and autonomous channels next.
VTP/1.0 · draft specification · open for comment
12 Get started
Put a padlock on every call.
No credit card · Sandbox in minutes · Talk to an engineer, not a deck