01
AI Capability
What the system can do: language, reasoning, code, tool use, long-horizon autonomy.
Method
The ATLAS model
A capability becomes risk when it acquires authority it was not designed to hold. The model traces that transition and can be read in either direction.
02
Exposure Surface
What the capability can reach: data, identity, tools, permissions, decisions.
03
Control Layer
What constrains it: identity-bound authorization, validation, approval, logging, classification.
04
Governance Layer
Who owns it, who accepts residual risk, and with what evidence.
Editorial governance
Editorial status
Every object carries a status: draft, under review, verified, published, superseded or archived.
Source policy
Primary sources first. Facts, ATLAS analysis, hypotheses and external claims are distinguished and attributed.
Verification
Framework pages expose official version, publisher, current status and last verification date.
Voice
Headlines may carry personality. Technical body copy stays evidence-based.
Automated checks
The build validates parity, links, sources and framework metadata before publishing.
How ATLAS works
See how ATLAS operates, how its changes are reviewed, and where human authority remains mandatory.
Read the operating model