Avoca — agentic threat model

Not certain from the listing — likely relies on commercial foundation models to generate customer service responses. Primary threats include prompt injection (allowing users to hijack the bot's behavior) and model hallucinations that could provide inaccurate business information.

Not certain from the listing — likely processes customer communication history and business FAQs. Threats include leakage of customer PII through conversational outputs and potential data poisoning if public-facing inputs are ingested into a RAG knowledge base without sanitization.

Not certain from the listing — likely uses a conversational orchestration framework to manage dialogue state. Threats include insecure tool integration if the agent is connected to booking systems or CRMs, allowing unauthorized state changes via natural language commands.

Not certain from the listing — presumably deployed as a cloud-hosted SaaS platform. Standard web application threats apply, including insecure API endpoints, lack of tenant isolation, and potential exposure of communication channel credentials (e.g., SMS/email API keys).

Not certain from the listing — no monitoring, logging, or guardrail mechanisms are detailed. A lack of real-time output filtering could allow toxic, off-brand, or malicious responses to reach customers undetected.

Not certain from the listing — no security certifications (such as SOC2) or compliance frameworks (such as GDPR/CCPA for customer data) are mentioned. Access control policies for managing business-side configurations are unspecified.

Not certain from the listing — there is no indication of multi-agent coordination or marketplace integrations. Risks are limited to standard third-party communication API dependencies (e.g., Twilio, email gateways).