Referent — agentic threat model

Not certain from the listing — likely relies on commercial closed-source LLMs to handle complex legal reasoning and document parsing. Primary threats include prompt injection altering legal interpretations or billing calculations, and potential data leakage of highly confidential client information to the model provider.

The agent manages highly sensitive legal data including client intake, matters, documents, and billing preparation. This creates severe risks of data exfiltration, unauthorized access to privileged attorney-client communications, and data poisoning of the firm's internal knowledge base.

Orchestrates multi-step workflows for routine operations (CRM, tasks, deadlines). Risks include tool misuse where the agent might incorrectly trigger emails, modify critical deadlines, or generate incorrect billing entries due to flawed planning or prompt injection.

Not certain from the listing — as a closed-source vertical legal SaaS, it likely runs in a cloud environment. Key threats include insecure API integrations with email servers and practice management databases, and potential privilege escalation if the agent's execution environment is compromised.

Not certain from the listing — requires robust logging and audit trails to track agent decisions, especially for billing prep and deadline modifications. A lack of observability could lead to undetected drift or silent failures in legal compliance workflows.

Features explicit 'lawyer-controlled approvals for critical actions' which acts as a vital human-in-the-loop (HITL) control. However, strict compliance with legal data protection regulations (e.g., HIPAA, GDPR) and attorney-client privilege boundaries must be enforced across all automated workflows.

Utilizes multiple 'AI agents for routine operational workflows' that must coordinate. This introduces risks of cascading failures or trust abuse if one agent (e.g., intake) passes compromised or malicious data to downstream agents (e.g., billing or document generation).