On 23 September sgit.ai opened a partnerships section and put one case in it: a partnership with UK Sovereign AI, the government-backed venture fund. Over the afternoon the page was released three times, and each release made it more careful about what it is. This piece walks through what the page says, what it proposes, and what it says it cannot show.
A method before a pitch
The partnerships index↗ sets out how every page in the section is written: start with what the other side is trying to do, use public material only, point every claim at evidence, state the fit precisely, and mark what is not public. Nothing on the pages should need permission to publish, and they are written to be forwarded (partnerships index↗). The index as snapshotted on 24 September also lists cases added later; on 23 September the section launched with the Sovereign AI case alone (version log, v0.5.7↗).
What the page says about the fund
The page describes the fund's mission from its own site and public reporting: early-stage equity, five frontiers of which one is trust, safety and assurance, and an R&D procurement scheme whose four initial challenge areas include "safe AI agent adoption for cybersecurity professionals" (the proposal↗). It quotes the fund's idea of "a sovereign edge" rather than a sovereign wall (the proposal↗). The test the page picks out is anchoring: a company must have, or commit to building, a significant and enduring UK presence (the proposal↗).
Where it says the two meet
A fit table maps each stated priority to vaults that can be opened with a read key today (the proposal↗). The safe-agent-adoption challenge area maps to Licence to Operate, Agentic Browser Isolation, the agent permission games and the Risk Mandate vault. The assurance frontier maps to the AIUC-1 conformance vault and a penetration test delivered as a vault. Citable regulation maps to the DSIT AI Risk Toolkit vault, Regulation Graph and the GDPR Standards Atlas (the proposal↗). Compute efficiency is described as "Not our frontier, but a measured property of the substrate" (the proposal↗).
Three shapes, all proposed
The page proposes three partnership shapes: a procurement prototype for the safe adoption of agents, official guidance and regulation published as citable graphs, and vaults as a shared substrate for the fund's portfolio (the proposal↗). It says it would start with the first, because a public simulation already exists and the gap to a real prototype is smallest (the proposal↗). None of these is a product or an agreement. The page says so: "There has been no conversation with the fund; this is the case we would make if there were one." (the proposal↗).
Anchoring, in the founder's words
Release v0.5.8 added a section written in the first person, "because only the founder can supply it" (version log↗). It states thirty years in the UK as practitioner, CISO and founder, one UK exit, and The Cyber Boardroom Limited, UK-registered and trading for almost two years, as the commercial vehicle behind sgit.ai and RiskMandate.ai (the proposal↗). The same facts went into one row of the About the author↗ page (version log↗). The section also argues that a closed champion is one acquisition away from not being sovereign, while an open-source substrate cannot be bought out from under the country that runs on it (the proposal↗). The page's own limits section says anchoring is "stated, not yet documented here" (the proposal↗).
A title that says what it is
Release v0.5.9, under an hour later by the git log, changed the title. The log's reason: "A reader landing from a forwarded link had no way to know this was a proposal from sgit.ai's side." (version log↗). The page now opens: "This is a proposal from our side, published in the open." (the proposal↗).
What the page says it cannot tell them
The closing section lists its own gaps: anchoring is stated rather than documented, sgit holds no compliance certification of any kind, the investment ask is left empty, and the fit is partial (the proposal↗). In its words: "Of five frontiers we serve one directly and touch one other." (the proposal↗).