Scope
Virturing can connect automated agents, customer-owned AI workers, deterministic workflows, and human teams to telephone calls. The customer deploying a workflow controls its purpose, audience, instructions, knowledge, tools, disclosure, retention, and escalation design.
Clear identity
Use an accurate organization and agent identity. Disclose automated participation when required by law, policy, sector rules, or reasonable caller expectation. Do not use a synthetic voice or script to deceptively impersonate a real person, authority, emergency service, or trusted institution.
Purpose and consent
Define the call purpose before processing contact data. Establish the lawful basis, required notice, consent, calling window, and opt-out method for each market and audience. Do not repurpose call content or contact lists incompatibly without a new lawful basis and notice.
Knowledge and prompts
Use reviewed sources appropriate to the task. Version prompts and knowledge, identify uncertain or time-sensitive content, and prevent the agent from inventing policy, pricing, eligibility, legal, medical, or account facts. Make the human path visible when information is missing or disputed.
Tools and actions
Give agents only the tools and credentials required for the job. Validate inputs, constrain amounts and destinations, use confirmation for consequential actions, log tool results, and ensure a person can reverse or review actions where appropriate.
Human escalation
Every deployment should define conditions that require a person, including caller request, low confidence, distress, vulnerable status, protected or sensitive data, disputed identity, tool failure, policy boundary, high value, and potential harm. Transfer context without forcing the caller to repeat the full interaction.
Crisis and self-harm response
Treat statements suggesting self-harm, suicide, violence, or immediate danger as a high-severity boundary. The agent must not continue an ordinary sales, collections, support, or advice script. Configure it to avoid harmful instructions, state its limits, offer locally appropriate emergency or crisis options, and transfer to a trained human where that route is available. Test these paths before launch and review failures as safety incidents.
Consequential decisions
Do not use Virturing as the sole decision-maker for diagnosis, treatment, employment, housing, education, credit, insurance, legal status, public benefits, law enforcement, or similarly consequential rights. Qualified people must make or meaningfully review those decisions under applicable law.
Recordings and transcripts
Enable recording and transcription only for a defined purpose with required notice, consent, access controls, and retention. Treat transcripts and summaries as derived records that can contain errors. Provide a route to correct important information and preserve the original call context where appropriate.
Testing and monitoring
Test representative accents, languages, noise conditions, interruptions, edge cases, tool failures, opt-outs, and escalation before launch. During operation, review quality, latency, errors, complaints, unexpected tool use, bias signals, caller confusion, and unresolved handoffs.
Security and data minimization
Limit prompts, context, logs, and model-provider data to what the call needs. Do not place passwords, full payment credentials, or unrelated sensitive data into an agent context. Use scoped credentials, bounded tool permissions, short-lived media tokens, and documented deletion paths.
Reporting concerns
Customers should provide a complaint and correction path to callers. Security concerns can be sent to security@virturing.ai, suspected abuse to abuse@virturing.ai, and privacy requests to privacy@virturing.ai. Include the relevant number and approximate call time without sending unnecessary sensitive data.