Enjo AI — agentic threat model

Not certain from the listing — The specific foundation models powering the 'enterprise-grade AI platform' are not disclosed. Threats include prompt injection attacks that could bypass customer service guardrails to output inappropriate or malicious content.

The agent is trained on company knowledge, tickets, and apps. This creates a high risk of data poisoning if malicious actors submit tickets designed to corrupt the agent's knowledge base, or data exfiltration of sensitive customer PII via engineered prompt queries.

Enjo uses 'AI flows and AI actions' for complex integrations. Framework-level threats include insecure tool execution where an attacker manipulates the agent into executing unauthorized API calls or routing tickets to malicious external endpoints.

Not certain from the listing — The hosting environment, sandboxing of 'AI actions', and secret management for Slack/Teams API tokens are not detailed. Compromise of these secrets could lead to unauthorized workspace access.

Features advanced analytics for performance, knowledge gaps, and sentiment analysis. However, it lacks explicit security-focused observability, such as real-time prompt injection detection or automated guardrail monitoring.

Not certain from the listing — While described as an 'enterprise-grade AI platform', specific compliance certifications (e.g., SOC 2, ISO 27001), data retention policies, and role-based access controls (RBAC) are not explicitly detailed.

Operates within a collaborative ecosystem (Slack, Teams, and connected apps). The primary threat is trust abuse, where the agent is used as a vector to launch social engineering attacks or distribute malicious payloads to internal employees.