AI Disclosure

Our AI agent is named the agent. It is software, not a person.

the agent runs on a combination of language-model, text-to-speech, and speech-to-text providers — all US-based, all on standard API tiers contractually prohibited from training on your call data. The full processor list (current vendor names + what each one does + where each operates) is in section 5 of the Privacy Policy. We keep vendor names there, not here, so we can swap providers without rewriting this page.

What the agent does not do — voice cloning of callers, speaker identification, biometric voice analysis — is covered in section 3 (what data the AI sees) and section 6 (no voiceprints).

Inbound calls (someone calls your business):

• Answers within a couple of rings.
• Greets the caller using the greeting wording you set up.
• Asks what the caller needs.
• Takes a message, books an appointment on your calendar if you’ve connected one, or escalates to you if the situation calls for it.
• Sends you the transcript and a short summary you can act on.

Outbound tasks (you ask the agent to make a call):

• You go to your Tasks dashboard, write “call Jorge at 555-0123 about the Tuesday delivery.”
• the agent places the call and runs the conversation, identifying as an AI by default.
• You get the transcript and outcome.

That’s the product. The agent is not a personal assistant for your life, not a sales auto-dialer, not a 911 service, and not a substitute for a doctor, lawyer, or accountant.

• Call audio — in real time while the call is happening, and stored if recording is enabled for that call or account.
• Transcripts — every call is transcribed. That’s how the AI works; it operates on text. Transcripts are stored.
• Your account context — your business name, hours, custom instructions for the agent, and your calendar availability if you’ve connected one.

What the AI does not see:

• No voiceprints. We do not create or store a unique voice fingerprint per caller. This is an architectural commitment — there is no setting to turn it on. See section 6.
• No speaker identification. We don’t “recognize” callers by voice across calls.
• No biometric voice analysis — no emotion-detection, no health-screening from voice, no age estimation.

The full processor list and what each one sees lives in the Privacy Policy.

Both of the following are always on. Neither can be disabled by an account holder.

Floor #1 — truth when asked. The agent will answer truthfully if a caller asks whether they’re talking to a person. If a caller says “are you a real person?” or “is this a bot?” or any variation, the agent says clearly that it’s an AI. This is enforced through the agent’s locked system instructions on every call.

Floor #2 — proactive AI identification in the greeting. Every greeting includes an AI identification (e.g., “this is the agent, an AI assistant for [your business]”). You can customize the greeting wording — the agent’s name, the tone, the opening question, the business reference — but the AI identification stays. We don’t let it be stripped, and the editor surfaces a warning if you try.

So the posture is: every caller hears that they’re talking to AI in the opening line, AND has a clear way to confirm it any time during the call by asking. This goes beyond the California SB 1001 / Texas TRAIGA truth-when-asked minimum and addresses the more recent state AI-disclosure laws too.

One disclosure, everywhere. We apply the same disclosure on every call, nationwide. See section 5.

We don’t run per-state disclosure rules. We apply one uniform disclosure on every call, nationwide. The greeting always identifies the agent as AI and states that the call may be transcribed — and adds “and recorded” when recording is enabled for your number.

The default greeting is something like: “Hi, this is the agent, an AI assistant for [your business]. This call may be transcribed.” With recording on, it becomes “…This call may be transcribed and recorded.”

Applying this to every call, rather than only to callers in certain states, is deliberate: it meets or exceeds the requirements of the all-party-consent recording states and the state AI-disclosure laws (such as California SB 1001 and AB 2905) without us having to guess a caller’s location. The AI-identification part of the greeting is locked (Floor #2 in section 4) and the truth-when-asked floor (Floor #1) stays on no matter what — neither can be removed from the greeting or the outbound-task form.

Call My Agent does not collect, store, or process voiceprints, speaker-identification fingerprints, or biometric voice analysis of callers.

This matters most in Illinois, where the Biometric Information Privacy Act (BIPA) creates a private right of action with statutory damages of $1,000 (negligent) or $5,000 (intentional) per violation for businesses that collect biometric identifiers without specific written consent. Our architecture forecloses that exposure cleanly: we don’t collect those identifiers in the first place.

If we ever offer a feature where the account holder’s own voice is cloned for use as the agent voice, we’d build a separate consent flow per ElevenLabs’s voice-cloning terms — and the cloned voice would be the account holder’s, never a caller’s.

Recording is off by default for your business phone number. Transcripts are still saved (the AI runs on text), but the underlying audio is not stored unless you turn recording on.

Turning recording on or off is a single decision per phone number. You flip it on for the whole line when you need to replay actual voice (training, dispute resolution, accuracy verification). You leave it off if transcript-search is enough. No per-call toggle in v1.

When recording is on, the agent’s greeting includes a “this call may be recorded” notice at the start of every call. The recording stream doesn’t start until that notice has played, so a caller who hangs up during the announcement is never recorded.

When recording is on, the notice is combined with the AI-identification phrase in the same greeting, and that combined greeting plays on every call, nationwide. Applying it everywhere — not just to callers in certain states — satisfies both state recording-consent law (including the all-party-consent states) and state AI-disclosure law (see section 5).

Outbound calls inherit the recording setting of the originating business number — if your inbound is recorded, your outbound is too. The same combined notice plays at the start.

• We don’t train our own models on it. Your transcripts, call audio, and custom instructions do not enter any training set we control.
• We don’t train our processors’ models on it either — Anthropic and OpenAI’s standard API tiers contractually prohibit training on customer data. ElevenLabs’s business tier we use is on the same posture. If any of those change, we update this page.
• We don’t sell it. No data brokers in our supply chain.
• We don’t share with anyone outside the processors listed in the Privacy Policy.

the agent makes mistakes — sometimes mishearing a name, sometimes summarizing a call in a way that misses nuance, sometimes booking the wrong time slot if the caller was ambiguous. The binding legal framing (warranty disclaimer, your responsibility to verify, the agent doesn’t sign contracts on your behalf) is in the Terms of Service, section “Responsibility for AI output.” This section is the operational counterpart: what you can actually do about a bad call.

What you can do:

• Review the transcript — every call. The summary is a first pass.
• Verify critical info — appointment times, prices, contractual commitments — before acting.
• Correct the agent. Mark a call as “the agent should have done X.” Your correction feeds back into your account’s persona settings so the next similar call goes better.
• Ask for a human review of a specific call through your account dashboard. We’ll work through it with you.

the agent is built to know when it’s out of its depth. When the agent is uncertain, when a caller is upset, or when the conversation needs judgment the agent doesn’t have, the agent escalates to you instead of guessing.

What escalation looks like:

• In-call. the agent can offer to take a detailed message and have you call back, can offer to text you mid-call so you can join if you’re free, or can transfer the call to a number you’ve set up in your account. The exact behavior depends on your settings.
• After the call. Every call ends with a transcript and a summary in your dashboard. If the agent flagged the call as “needs your follow-up,” you’ll see that flag and a suggested next step.
• Push, SMS, and email notifications. You pick which channels you want, in your account settings. Emergencies get the loudest channel by default (see the 911 / Emergency Handling Policy).

What you can configure:

• Trigger words and escalation rules — caller phrases or topics that should always go to you, not the agent.
• Off-hours behavior — what the agent does outside business hours (take messages by default; escalate certain callers immediately if you set them as VIPs).
• Per-caller handling — known contacts (suppliers, key customers) can be routed differently than unknown numbers.

What the agent won’t do:

• Pretend it tried to reach you when it didn’t. If escalation fails (your phone is off, your inbox is full, your forward number doesn’t pick up), the agent tells the caller honestly and offers an alternative (take a message, suggest they try another contact method).
• Hand the caller off to someone else’s queue. the agent doesn’t transfer to a third-party call center on your behalf. You configure your own forwarding targets.

• Not for medical, legal, or financial advice transmission. Don’t have the agent read prescriptions to patients or quote insurance terms to clients. It’s not built for that and we’re not licensed to support it.
• Not for 911 or emergencies. If a caller indicates an emergency, the agent tells them to hang up and dial 911 directly. See the 911 / Emergency Handling Policy.
• Not for marketing auto-dialing. Outbound tasks are one-at-a-time and require a reasonable business basis for the call.
• Not for hard problems beyond the agent’s current capability. Heated emotional callers, complex multi-step negotiations, or calls that need real human judgment — the agent is built to escalate to you when it’s uncertain.

Call My Agent is a tool. Your business is the one operating the line — you’re the “caller” or the “operator” under the laws below. Your use of the service has to comply with them. What we ship by default is built to help; the Permitted use section asks you to keep it that way and not configure around the floors.

• California SB 1001 — requires businesses to disclose use of bots in some commercial / political contexts. Our truth-when-asked floor plus the default greeting we ship cover the typical use case. If you customize the greeting in a way that strips proactive disclosure, the floor still catches it, but you’re responsible for whether your overall posture meets SB 1001 for what you’re using the agent for.
• California AB 2905 — outbound AI voice calls to California recipients must disclose AI at the start. The uniform disclosure we apply on every call includes that AI identification, so California recipients get it automatically; it can’t be stripped.
• Texas TRAIGA (Responsible AI Governance Act, effective January 2026) — disclosure when a person might reasonably believe they’re talking to a human. The disclosure regime we ship covers this; your customization choices are yours.
• Illinois BIPA — no biometric identifiers (including voiceprints) without written consent. We don’t collect voiceprints (section 6), so this exposure is foreclosed on our side. If you ever clone your own voice for the agent voice in a future feature, BIPA-style consent flow applies on your side and we’d build the tooling for it.
• Federal TCPA + state robocall laws — these regulate the caller. You’re the caller. We provide DNC checks, time-of-day windows, attestation flows for outbound; the Permitted use section prohibits TCPA-violating uses. False attestation by you is a terminable violation.
• All-party-consent recording states (CA, IL, FL, MD, MA, MT, NH, PA, WA) — when recording is on for your number, our uniform greeting includes a recording notice on every call, so callers in these states are covered automatically.

Questions about how the AI works, or about anything on this page:

• Email: privacy@callmyagent.ai

When this page changes in a material way — new processors, changed defaults, new state coverage — we update the “Last updated” date and notify active accounts.