Agent conduct addendum

1. Scope

For purposes of this addendum, an "Agent" is any software system that takes autonomous action on the Service on behalf of Customer — including agents that rely on large language models, decision policies, retrieval pipelines, or any combination of automation that produces outbound communications, message sending, number provisioning, or other state changes without per-action human approval. An "Operator" is the human or organization responsible for an Agent's deployment, scope, and supervision.

2. Identity

Every Agent that operates on Dials must carry a Dials seat with an OAS DID bound to it, and that DID must trace, through Customer's organizational lineage, back to a real human Operator who accepts responsibility for the Agent's actions. Agents that cannot be bound to a real human cannot be deployed on the Service. The act of provisioning an Agent on Dials is an affirmation that the Operator has the authority to accept that responsibility.

3. Scope and capability

Operators must grant Agents the minimum scope necessary for the work they perform. Read scopes are granted before dial scopes; dial scopes are granted before send scopes; and any scope that places Agents in communication with minors, healthcare patients, financial-services customers, or any of the Special Categories listed in the AUP requires pre-approval from legal@dials.com.

Operators must implement reasonable rate, volume, and frequency controls on Agents and must monitor the Agent's behavior for drift, hallucination, or scope escalation. Dials will refuse to execute, and will record as denied, any action an Agent attempts that exceeds its scope.

4. Disclosure

Where required by law, and in any communication in which a reasonable recipient is likely to assume they are speaking with a human, Agents operating on Dials must disclose, in clear and conspicuous terms, that the recipient is communicating with an automated system. Specifically, Agents must comply with applicable state law (including California's Bolstering Online Transparency Act and similar statutes), the FTC's guidance on AI disclosure, and any FCC orders concerning AI-generated voice.

Agents must not impersonate a specific real person — living or dead — without that person's documented, revocable consent. Agents must not generate the voice of a specific real person without that person's documented, revocable consent.

5. Consent handling

Agents must treat all consent state as authoritative. An Agent must not place a call or send a message that requires consent under applicable law unless the system has a recorded, dated, source-attributed consent for the recipient that has not been withdrawn. An Agent that hears, reads, or otherwise receives a request to opt out, stop contact, or be removed from a list must, within the time required by applicable law and in no case longer than ten (10) business days, register that opt-out across all channels and seats Customer operates.

Agents must recognize ambiguous opt-out language ("please stop calling," "remove me," "I'm not interested," and equivalents in the recipient's language) and must err on the side of honoring the request. Failure to honor opt-outs is a material violation of this addendum and the AUP.

6. Recording and emotion

When an Agent participates in a recorded call, the Agent must play the configured recording disclosure before any substantive exchange and must not arm the recorder until the recipient has acknowledged. Agents must not attempt to discourage, distract, or talk over recipients who decline recording.

Agents must not deploy techniques designed to manipulate a recipient's emotional state for the purpose of overriding the recipient's interests — including, without limitation, synthetic urgency, manufactured grief, simulated romantic affection, and prolonged social-engineering exchanges. These are violations of this addendum and may also violate consumer-protection law.

7. Honesty and refusal

Agents must not make false statements about themselves, the Operator, the products they are selling, the regulatory status of those products, or the consequences of declining to engage. Agents must not threaten recipients, including with the loss of benefits the recipient is in fact entitled to.

Operators must equip their Agents with the ability to refuse a request — from the recipient, from a co-Agent, or from a human supervisor — when complying would violate the AUP, this addendum, applicable law, or the Operator's published policies. The refusal itself is recorded.

8. Audit

Every Agent action on Dials writes a usage event that includes the Agent's DID, scope, decision rationale where the Operator chooses to record it, and the resulting disposition. Operators must retain Agent prompts, system instructions, and tool invocation logs for the period required by applicable law and for the period needed to investigate complaints — in no case less than ninety (90) days.

9. Human supervision

Operators must maintain meaningful human supervision of their Agents. "Meaningful" means that a named human Operator (a) has authority to suspend the Agent immediately, (b) receives notice of unusual behavior, (c) reviews a representative sample of Agent interactions on a regular cadence, and (d) responds to recipient complaints within a commercially reasonable time. Operators must publish a means for recipients to report Agent misconduct and must respond to such reports.

10. Enforcement

Violations of this addendum carry the same consequences as violations of the AUP, up to and including immediate termination, preservation of evidence, and notification of impacted carriers, regulators, and law enforcement. Dials may impose interim controls — including reduced scope, throttling, and pre-approval requirements — on Agents that show patterns of risky behavior.