UseJunior AI Assistant — agentic threat model

Not certain from the listing — The underlying foundation models are closed-source and unspecified. Threats include adversarial prompt injection embedded within uploaded contracts designed to hide malicious clauses or manipulate the review output.

Not certain from the listing — The agent processes large documents (up to 150 pages) and performs comparisons, likely utilizing a temporary vector store or RAG pipeline. Threats include data exfiltration of highly sensitive legal agreements and potential data leakage across parallel tenant requests.

Not certain from the listing — Orchestration handles document parsing, parallel execution, and email generation. Threats include insecure parsing of malicious PDF/Word attachments and prompt injection exploiting the email-to-agent pipeline.

Not certain from the listing — The infrastructure relies on email servers (SMTP/IMAP) and cloud document processing. Threats include interception of unencrypted emails, lack of transport layer security (TLS) enforcement, and container escape during document parsing.

Not certain from the listing — The system provides verifiable citations to mitigate LLM hallucination, but automated monitoring of model drift or adversarial inputs is not detailed. Threats include blind spots where the model misses critical clauses without triggering alerts.

The agent is explicitly ISO 27001 certified, indicating established information security management systems. However, relying on email as the primary interface introduces significant identity and authorization risks, such as email spoofing (SPF/DKIM/DMARC bypass) leading to unauthorized document access.

Not certain from the listing — The agent operates as a standalone vertical solution via email. Threats are primarily external, such as phishing campaigns impersonating the 'Junior' assistant to harvest sensitive legal documents from lawyers.