Conveyor — agentic threat model

Not certain from the listing — likely relies on commercial LLMs (e.g., GPT-4 or Claude) via API. Threats include prompt injection bypassing guardrails to leak sensitive security documents or past questionnaire answers.

Integrates heavily with internal wikis, Confluence, and past questionnaire databases. Threat of RAG poisoning (e.g., malicious content in Confluence leading to incorrect security answers) and data exfiltration of sensitive NDA-gated documents.

Orchestrates multi-step tasks (reading questionnaires, querying RAG, updating Salesforce/Jira/Zendesk). Threat of tool misuse where a malicious questionnaire or prompt triggers unauthorized updates to systems of record.

Not certain from the listing — likely hosted as a multi-tenant SaaS. Threats include container compromise, insecure storage of API keys for Salesforce/Jira/Slack, and lack of network isolation when accessing customer portals.

Features 'no-code guardrails' to control behavior based on business rules (e.g., deal size). Threat of guardrail bypass via adversarial prompt injection in questionnaires, and potential logging blind spots regarding what data is sent to external LLMs.

Not certain from the listing — as an enterprise security tool, it likely undergoes SOC 2 audits and implements RBAC, but specific compliance certifications and credential management practices are not detailed in the public directory.

Connects directly to communication channels (Slack, Teams, email) and enterprise systems (Salesforce, Jira, Zendesk). Threat of cascading failures or unauthorized actions if a compromised upstream agent or user interacts with Sue via Slack.