Digitar AI — agentic threat model

Not certain from the listing — Uses closed-source speech-to-speech and LLM models to drive conversational experiences. Threats include voice-based adversarial prompt injection, model reprogramming, and potential output manipulation during customer support interactions.

Not certain from the listing — Processes customer transcripts and lead generation data. Threats include unauthorized access to stored conversation logs, data exfiltration of personally identifiable information (PII), and lack of clarity on data retention policies.

Not certain from the listing — Uses a proprietary no-code orchestration framework to optimize conversational flows. Threats include prompt injection bypassing the intended use case (e.g., turning a lead-gen bot into an arbitrary text generator) and insecure handling of session state.

Not certain from the listing — Deployed via a webclient and hosted platform. Threats include web application vulnerabilities, insecure streaming protocols for speech-to-speech, and potential container/hosting infrastructure compromise.

The platform explicitly features advanced monitoring, analytics, and detailed transcripts for each conversation, which significantly aids in detecting anomalies, drift, and malicious inputs in real-time.

Not certain from the listing — No explicit security certifications (e.g., SOC2, ISO) or compliance frameworks (e.g., GDPR for voice recording consent) are detailed. Threats include regulatory non-compliance regarding voice data storage.

Not certain from the listing — Operates as a single-agent creation platform. Threats include malicious actors creating rogue assistants to harvest credentials or scam users under the guise of legitimate customer support.