Technical paper / Version 1.0
ContextOne Security.
An implementation review of ContextOne security architecture, covering agent authority, access controls, isolation and auditability.
Overview
Useful agents need access to information, tools and execution environments. ContextOne is designed to make those capabilities available through governed interfaces, with controls over which resources an agent can reach and which actions it can perform.
This paper explains the security architecture and the controls available to organisations deploying ContextOne. It is an implementation review, dated 30 October 2026, rather than an independent audit or certification.
What the paper covers
Bound agent authority
A model output is not an authorisation decision. The paper describes explicit permissions and controls for tools, structured data, ontology resources and unstructured information.
Contain powerful capabilities
Sandbox isolation and governance cover agent use of Python, Bash and command-line utilities, while service communication and encryption support the wider architecture.
Support review and investigation
The paper covers auditability, runtime guardrails, rate limits and cost controls, along with the organisation’s own access policies and operational procedures.
How to read the findings
The paper covers foundational, sandbox, ontology, tool-interface, structured-data, LLM-interface and Data Room security. Deployment settings and the organisation’s source-system permissions remain important to the resulting security posture.