Auxi — agentic threat model

Not certain from the listing — likely relies on third-party commercial foundation models. Main threats include prompt injection to bypass HR policy guardrails and potential leakage of sensitive employee PII through model outputs.

Ingests 'Existing Wiki' data and connects to HR databases. Highly vulnerable to RAG data poisoning (e.g., an attacker editing a wiki page to inject malicious instructions) and unauthorized data exfiltration of sensitive corporate knowledge.

Orchestrates workflows across '100s of pre-built integrations'. Insecure tool calling is a major threat, where a manipulated user query could trigger unauthorized actions (e.g., modifying employee records or triggering external API calls) without proper validation.

Not certain from the listing — presumably hosted as a multi-tenant SaaS platform. Key threats include insecure storage of API credentials for the 100s of integrated apps and potential tenant isolation failures.

Not certain from the listing — no mention of built-in guardrails, audit logging, or drift monitoring. Lack of observability could allow stealthy prompt injection attacks or data harvesting to go unnoticed.

Handles HR and employee self-service, demanding strict access control. The primary threat is privilege escalation, where the agent fails to properly map Slack/Teams user identities to their corresponding HR database permissions, leaking restricted data.

Integrates directly with Slack, Teams, and numerous third-party apps. This creates a large attack surface where a compromise in a connected third-party tool could cascade into Auxi, or Auxi could be used as a vector to attack other enterprise systems.