Nuance Mix — agentic threat model

Uses a hybrid model of traditional NLU, ASR, TTS, and LLM enhancements. Threats include adversarial voice/text inputs (prompt injection) bypassing NLU intent boundaries, and potential misalignment of the integrated LLM components.

Not certain from the listing — the platform processes conversational data and integrates with enterprise databases, but specific details on vector databases, RAG pipelines, or training data ingestion are not detailed. Standard risks include data exfiltration of customer PII during dialog sessions.

Orchestrates dialog and bots using DIY tooling and APIs. Vulnerabilities here involve insecure tool integration with CCaaS platforms (Genesys, Avaya) and potential manipulation of the dialog state machine by malicious inputs.

Deploys within Microsoft Azure environments and integrates with external CCaaS stacks. Threats include Azure misconfigurations, insecure API endpoints connecting the IVR to telephony infrastructure, and unauthorized access to orchestration APIs.

Provides built-in testing and analytics tools. Risks include logging sensitive customer data (such as credit card numbers or health records) in plaintext within the analytics logs, or failing to detect drift in LLM-enhanced dialog paths.

Explicitly designed for regulated industries (Healthcare, Banking, Telecom) with governance controls. The primary threat is compliance failure (HIPAA, PCI-DSS) if LLM-driven features generate unapproved or non-compliant responses that bypass standard governance filters.

Not certain from the listing — while it integrates with CCaaS ecosystems and omnichannel environments, there is no explicit mention of an autonomous multi-agent marketplace or dynamic agent-to-agent trust delegation.