VoAgents — agentic threat model

Not certain from the listing — likely relies on external LLMs combined with Speech-to-Text (STT) and Text-to-Speech (TTS) engines. Key threats include voice-based prompt injection (jailbreaking via spoken audio) and adversarial audio inputs designed to manipulate agent behavior.

Not certain from the listing — requires access to customer data and business context to handle sales and support calls. Threats include unauthorized exfiltration of CRM data during a call session and poisoning of the knowledge base used to answer customer queries.

Not certain from the listing — orchestrates dialogue flow, calendar booking, and CRM updates. Insecure tool integration could allow an attacker to manipulate API calls to the CRM or calendar via malicious spoken instructions during a call.

Not certain from the listing — requires hosting voice-streaming infrastructure (e.g., WebRTC, SIP) and API integrations. Threats include SIP trunk abuse, toll fraud, and denial-of-service attacks targeting the real-time voice channels.

Not certain from the listing — requires robust call transcript logging and audio recording. A major threat is the accidental logging of sensitive PII or financial data (e.g., credit card numbers spoken by customers) in plaintext logs.

Not certain from the listing — must navigate strict telecommunication regulations (e.g., TCPA in the US, GDPR for call recordings). The listing does not specify compliance frameworks, encryption standards, or consent-handling mechanisms for outbound dialing.

Not certain from the listing — primarily functions as a point-to-point integration agent (Voice to CRM/Calendar) rather than participating in a complex multi-agent marketplace, minimizing cascading multi-agent trust risks.