Context — agentic threat model

Not certain from the listing — The specific foundation models powering the Context Engine are undisclosed. Standard risks include prompt injection bypassing workspace boundaries and adversarial inputs poisoning the personalized drafting outputs.

The Context Engine continuously learns from all user data, past work, files, chats, and tasks. This creates a massive attack surface for data/knowledge-base poisoning, where malicious documents or chats injected into the workspace could corrupt the RAG pipeline or lead to unauthorized data exfiltration via universal search.

The agent automates tasks, drafts documents, and integrates with external tools. Insecure tool integration and memory poisoning are critical threats here, as malicious inputs stored in persistent memory could trigger unauthorized tool execution or manipulate automated workflows.

Not certain from the listing — While end-to-end encryption is mentioned, the underlying hosting, containerization, and sandboxing of document generation processes (especially spreadsheets/macros) are not detailed, leaving potential gaps for host compromise or lateral movement.

Not certain from the listing — There is no mention of evaluation frameworks, real-time monitoring, guardrails, or logging mechanisms to detect anomalous agent behavior or drift in the Context Engine.

The listing highlights 'end-to-end encrypted for secure and private work' as a primary security control. However, details regarding access control (RBAC), identity management, and compliance certifications (e.g., SOC2, ISO) are not specified.

Not certain from the listing — While the platform supports real-time collaboration and team chat, it is unclear if it supports autonomous multi-agent coordination or third-party agent marketplace integrations that could introduce cascading trust failures.