The provisions this supports, named one by one.
Provision mappings are WestGate Data Science analysis. Select a mandate to see the detail.
What an assessor can verify without taking your word for it.
Four properties of the record itself, which is what an assessor tests. These are the properties the Part 500 certification rests on, and they apply to every mandate above.
{{ s.v }}
Which controls this actually maps to, and which it does not.
Where Churchill technically supports specific provisions within each mandate. It strengthens your evidence for these areas; your program covers the rest. Every mandate below lists what Churchill does not do for it, at the same weight as what it does. SWIFT CSP is open below; select any of the others to read its detail.
Who can stop a runaway workload in your environment, and prove it?
The AI Kill Switch Act, H.R. 9917, was introduced on 23 July 2026 by Representatives Lieu and Moran and referred to the House Committee on Homeland Security. Its obligations fall on model developers, not on financial institutions. We keep it off the mandate map for that reason: it does not bind you, and pretending otherwise would discount the regimes that do.
The question underneath the bill is not legislative. It is architectural, and it is already yours.
{{ a.v }}
Legislative status as introduced; the bill had not been enacted at the time of writing. Nothing here claims Churchill satisfies an obligation the act places on model developers.
Take it to your gap assessment.
The mappings above are our analysis, not a determination. Run them against your own control set, and test the evidence claim on a host you choose before anyone writes it into a program document.
Churchill is built and validated on IBM LinuxONE.