Jarni AI — agentic threat model

Not certain from the listing — likely utilizes third-party speech-to-text, text-to-speech, and LLM APIs. Primary threats include voice-based prompt injection (audio injection) causing the agent to deviate from scripts, leak system prompts, or output inappropriate content during live calls.

Not certain from the listing — ingests real-time voice streams, generates transcriptions, and syncs with CRMs. Threats include the accidental ingestion and storage of sensitive PII/PCI in call transcripts, and potential data exfiltration via unauthorized webhook payloads.

Orchestrates voice interactions, appointment booking, and CRM updates. Threats include insecure tool integration where an attacker manipulates the conversational flow to trigger unauthorized CRM writes, malicious call routing, or spam appointment bookings.

Not certain from the listing — likely hosted in a cloud environment to support API access and real-time webhooks. Threats include unauthorized API access, lack of rate limiting on voice endpoints leading to telephony denial-of-service (TDoS), and insecure webhook endpoints.

Features automated QA that reviews 100% of calls for script adherence and tone. Threats include evaluation gaming, where malicious actors or rogue agents manipulate voice tone and keywords to bypass compliance checks, and blind spots in detecting sophisticated prompt injections.

Not certain from the listing — handles voice recordings and customer PII, necessitating strict compliance with GDPR, CCPA, and potentially PCI-DSS. The listing does not specify encryption standards, access controls, or compliance certifications.

Integrates directly with external CRM ecosystems and webhook targets. Threats include cascading failures if downstream CRM APIs experience outages, and trust abuse where the receiving CRM implicitly trusts data written by the AI agent without secondary validation.