Responsible AI Policy
This Policy explains how Aspen Blue Capital LLC ("Aspen") understands the responsible use of artificial intelligence in AIBI and what we require of those who use it. It forms an integral part of the Platform Terms of Service and complements the Acceptable Use Policy.
AIBI builds agents that speak on the phone with a synthetic voice and can sound human. We believe that capability is only legitimate if the person on the other end is not deceived about what they are talking to. That principle governs this entire document.
1. Our principles
- Transparency: a person has the right to know they are talking to an automated system.
- Consent: a person's voice is not replicated without their express permission.
- Human control: decisions that significantly affect people require an accountable person.
- Proportionality: only the minimum data necessary for the purpose is processed.
- Accountability: whoever deploys the agent answers for what the agent does.
2. Disclosure: the agent must identify itself as AI
The Customer must configure its agents to disclose their artificial nature where the law requires it and, in any event, whenever the other party asks. Designing an agent to claim it is human, or to dodge the question, is prohibited.
As good practice we recommend disclosing it at the start of the conversation. Several jurisdictions require this disclosure expressly (for example, US state rules on bots in commercial interactions), and the regulatory trend is moving in that direction.
3. Synthetic voice and the TCPA framework
In February 2024, the FCC ruled that AI-generated voices are "artificial" voices for the purposes of the Telephone Consumer Protection Act (TCPA). The practical consequence is direct: automated calls with AI agents to US numbers are subject to the prior express consent requirement, with the exceptions provided by law, and breaching it exposes the caller to significant per-call penalties.
The Customer is responsible for holding that consent. Aspen does not verify it in advance, but may require evidence of it and suspend campaigns or accounts in accordance with the Acceptable Use Policy.
4. Voice cloning
Voice cloning is only permissible with strict safeguards. The Customer undertakes to:
- Hold the express, informed and verifiable consent of the voice owner, and keep evidence of it.
- Hold the necessary rights over the audio samples it uploads.
- Not impersonate any person or mislead anyone about the identity of the speaker.
- Not clone the voices of minors, nor of deceased persons without authorisation from the appropriate party, nor of public figures to simulate statements.
Cloned voices are private to the account that created them and are not shared with other customers or added to public catalogues. A voice may constitute biometric data subject to heightened regulation; the Customer must check the requirements of its jurisdiction before using this feature.
5. Call recording
The Platform can record and transcribe calls. The Customer must comply with the applicable recording regulations, which vary substantially: some jurisdictions require the consent of all parties (several US states) while others are satisfied with one. When in doubt, the safe practice is to inform and obtain consent at the start of the call.
6. Limits of the technology
Language models are probabilistic. Although we work to reduce errors, the Customer must assume that an agent may:
- Hallucinate: confidently state false or invented information.
- Misinterpret audio with noise, accents, crosstalk or poor line quality.
- Go off script in the face of unexpected input or deliberate manipulation attempts (prompt injection).
- Reflect biases present in its training data.
For that reason, agent output must not be treated as professional advice or as binding statements by Aspen, and must be validated before it produces material effects.
7. Uses that require human oversight
In accordance with the Acceptable Use Policy, the Platform must not be the sole basis for decisions with legal or similarly significant effects on a person (credit, employment, insurance, health, access to essential services). In those contexts there must be qualified human review and sector regulations must be met. The Platform is not an emergency service.
8. The person's right to end the conversation
The Customer must configure its agents to recognise and respect requests to end the call, not to be contacted again and to speak to a human being where that is required or reasonable, and must action those requests effectively in its own systems.
9. What Aspen does with conversation data
- Aspen processes recordings, transcripts and summaries on the Customer's behalf, as Processor, in order to provide the Platform (DPA).
- Aspen does not sell identifiable conversation data.
- Aspen does not train or fine-tune models with identifiable Customer data; to improve its models it uses only aggregated and irreversibly de-identified data.
- Aspen may generate and use aggregated and irreversibly de-identified data to improve its systems, models and user experience, and share it with its group and partners, without re-identifying any person.
- The third-party models involved (see Sub-processors) process data under contract and limited to the purpose of providing the service.
10. Reporting a problem
If a Platform agent behaved deceptively, impersonated a voice or produced a harmful outcome, write to us at business@aspenbluecapital.com. We investigate reports and may suspend agents or accounts in accordance with the Acceptable Use Policy.
11. How this Policy evolves
The regulation of artificial intelligence and synthetic voice is under active development. We will update this Policy as the applicable obligations and our features change, publishing the date of the last update.