Zhunsheng (准绳)

Zhunsheng (准绳) is an AI-assisted legal work platform for practicing lawyers in China, delivered as a multi-tenant SaaS application.

Lawyers describe a goal or hand over case materials; the platform runs an agentic workflow inside a sandbox and returns finished work products as files — bank-statement analyses, evidence indexes, drafted and redlined documents, legal calculations — each traceable to its sources.

Scope of this assessment: the Zhunsheng production platform, including its web application, agent execution engine and sandbox, document processing pipelines (OCR, Office formats, evidence binding), legal reference databases, and supporting infrastructure.

Security highlights

• Agent code executes in a namespace sandbox with no network access; filesystem writes are limited to an explicit allowlist.
• Four deterministic gates — identity, artifact integrity, provenance, and judgment egress — validate deliverables before they reach the lawyer. All fail closed.
• Case data is isolated per tenant at a single authorization choke point, guarded by a dedicated cross-tenant regression suite in CI.
• Uploaded evidence is immutable to the agent; deletion is a lawyer-only action with a recoverable stage.
• External transport requires TLS 1.2 or higher; production backups are AES-256 encrypted and restore-verified.