AI Disclosure Policy

1. What our voice agents are

An AiAnchor agent is an AI system built on Retell AI’s conversational voice platform. It answers a business’s inbound phone line, understands natural speech, answers questions from that business’s approved knowledge base, books appointments and records lead details. After the call, the recording is transcribed and analysed (using Retell post-call analysis and the Anthropic Claude API) to produce the summary and metrics the business sees in its portal.

2. The spoken disclosure at call start

Every agent we deploy identifies itself as an AI assistant, and states that the call is recorded, in the first moments of the call, in the language of the call, before any personal data is collected. Example script: “Hi, you’ve reached [Business]. I’m their AI assistant, and this call is recorded so the team can follow up. How can I help?”

This in-call spoken disclosure is our compliance mechanism for Article 50(1) of the EU AI Act (Regulation (EU) 2024/1689), which applies from 2 August 2026 and requires that people interacting with an AI system are informed in a way that is perceivable during the interaction itself. A note buried in terms and conditions does not satisfy this; the agent must say it. The recording announcement also satisfies Greek call-recording rules (Article 11, Law 3471/2006), which require participants to be informed before recording.

Clients may adjust wording (tone, branding) but may not remove either element. The disclosure cannot be disabled.

3. Human oversight

Agents operate within scopes their business owner configures and approves: what they may answer, what they must not discuss, and where they hand off. Businesses review calls, transcripts and summaries in their portal and can correct the agent’s knowledge base at any time. AiAnchor monitors deployed agents for failures. PLACEHOLDER: describe the concrete review cadence/alerting once confirmed.

4. Reaching a human

A caller can ask for a human at any point. Depending on the business’s configuration, the agent transfers the call, takes a callback request for a named person, or provides direct contact details. Agents are configured to recognise emergencies and complex requests and route them out rather than improvise. PLACEHOLDER: confirm transfer/callback capability per plan.

5. If a caller objects to AI or recording

A caller who does not want to talk to an AI or be recorded can say so; the agent offers the human path in section 4 and does not press on. For rights over data from a past call (access, deletion), the caller’s request goes to the business they called, which controls that data; we assist that business. Callers can also reach us at info@aianchor.online and we will route the request. PLACEHOLDER: confirm whether recording can be disabled mid-call on request; if not, the objection path is the human handoff.

6. What the AI does with call data

• Anthropic (Claude API, used for call analysis): under its commercial terms, API inputs and outputs are not used to train Anthropic’s models.
• Retell AI (voice platform): recordings, transcripts and logs are retained per the configured retention policy (configurable from 1 day to 2 years). PLACEHOLDER: Retell’s public documentation does not state a no-training commitment; we must verify this in Retell’s DPA before asserting it here.
• AiAnchor does not use caller data to train any models of its own.

7. Limitations, honestly

• The agent can mishear (accents, background noise) and occasionally answers imperfectly or not at all. It answers only from the business’s approved information, but AI-generated speech can still contain errors.
• It does not give medical, legal or financial advice, and it is not an emergency service; it routes such calls to humans.
• Prices, availability and commitments stated by the agent are confirmed by the business in the follow-up where they matter.

8. Regulatory framing

Under the EU AI Act our voice agents are AI systems intended to interact directly with natural persons (Art. 50(1)); they are not used for emotion recognition or biometric categorisation, and the post-call sentiment analysis assesses conversation tone for customer-service quality on behalf of the business, not to identify or infer emotions of identified individuals for the purposes of Art. 50(3). PLACEHOLDER: counsel to confirm this classification. AI-generated audio content in the calls is disclosed at call start (Art. 50 disclosure above).

9. Questions

About this policy or an experience with one of our agents: info@aianchor.online.