Zowie — agentic threat model

Not certain from the listing — Zowie uses Generative AI to automate support tasks, but the specific underlying foundation models (e.g., GPT-4, Claude, or proprietary LLMs) are not disclosed, leaving potential vulnerabilities to adversarial prompt injection or model-specific exploits unverified.

Not certain from the listing — The platform integrates with existing customer service platforms and learns from interactions, implying a RAG or vector database setup, but the exact data storage, ingestion pipelines, and protection against training data poisoning are unspecified.

Not certain from the listing — Orchestration is managed via the proprietary 'Zowie Decision Engine' to prevent hallucinations and automate complex workflows, but the underlying agent framework, tool-calling mechanisms, and memory sanitization techniques are closed-source.

Not certain from the listing — As a paid enterprise platform, it likely runs in a cloud environment, but details regarding tenant isolation, API security, secrets management, and sandboxing of execution environments are not provided.

Not certain from the listing — The platform 'enables feedback & training to ensure continuous improvement' and guarantees accuracy, but the specific observability stack, guardrails, and real-time anomaly detection mechanisms are not detailed.

Not certain from the listing — While targeted at enterprises, the listing does not explicitly detail compliance certifications (such as SOC 2, GDPR, or HIPAA) or specific role-based access control (RBAC) policies governing the agent's actions.

Not certain from the listing — Zowie integrates with external customer service platforms and proactively reaches out to customers, but it is unclear if it interacts with other autonomous agents or operates within a multi-agent ecosystem.