A Zero Trust architecture that lets AI act autonomously, within limits you control
Security controls must be verifiable and authorisation boundaries clear. We build security and compliance requirements into delivery, from data processing to agent execution.
From question answering to autonomous execution, capability goes up and controllability goes down; the gap is the risk exposure. The more an agent can do, the more it can do wrong. Zero Trust has to be designed in on day one. You cannot add a foundation after the house is built.
Security architecture
Automated compliance audit gateway
Connects data acquisition with data processing.
Private deployment
Integrated hardware and software with the sandbox runtime and security console built in.
Engineering
9Xin Iron Claw Security Engine
A Rust-based runtime and security control system for skills and harness (execution framework) components.
Sandbox isolation
Agent actions execute inside a sandbox, isolated from business-system permissions.
Prompt-injection defence
Instructions embedded in external content are isolated and identified: observed content is data, not a command.
9Xin Compliant Acquisition Engine
A non-intrusive, exception-driven engine for acquiring heterogeneous data, operating within compliance boundaries and connected to the audit gateway.
Filing information
Filing information
- Algorithm filing: 网信算备440309441584901250013号
Data sourcing and legal basis
Let’s start with a specific business need.
Tell us where AI has yet to work in practice: where the data is, who will use the system and who approves its outputs. In a workshop, we will work with you to define a delivery plan with verifiable results.