Denki — agentic threat model

Not certain from the listing — Denki is closed source and does not specify its underlying foundation models. Potential threats include adversarial prompt injection to bypass audit controls or misaligned outputs leading to false compliance reports.

Denki ingests highly sensitive financial data from ERPs, AuditBoard, and Workiva. Key threats include data poisoning of financial records to hide fraud, unauthorized data exfiltration of sensitive financial/PII data, and lineage/provenance gaps in evidence collection.

The agent orchestrates complex workflows like walkthrough interviews, control testing, and financial reconciliation. Threats include insecure tool integration with ERPs, tool misuse (e.g., unauthorized transactions or data deletion during reconciliation), and memory poisoning.

Not certain from the listing — The hosting environment (SaaS, cloud, or on-premise) and sandboxing mechanisms for executing integrations with ERPs are not specified. Threats include container compromise, privilege escalation, and exposed API keys for ERPs.

Denki provides 'full audit trace documentation' and 'continuous monitoring', which implies some level of logging and observability. However, blind spots in LLM-based walkthrough interviews or evaluation gaming during control testing remain significant threats.

While designed to automate compliance (SOX 404, BSA/AML), the listing does not explicitly detail Denki's internal security controls (like SOC2, RBAC, encryption). Threats include unauthorized access to compliance reports and lack of strict identity/authorization controls over ERP integrations.

Denki integrates with enterprise platforms like AuditBoard and Workiva. While it acts as a specialized workflow agent, there is no explicit mention of multi-agent collaboration or marketplace interactions, limiting immediate A2A trust abuse threats.