The money reached the company. And the government paid the company to help design its evaluations. We have recovered Good Ventures’ filed payment and grant-oversight records for Pattern Labs, now Irregular, plus a third UK government contract. The new contract pays for work upstream of a score: choosing important capabilities, designing the evaluations and explaining why those tests cover the threat.
Explore the funding and contract map, open Irregular’s profile, or read the earlier contract investigation.
From grant announcement to reported payments
Good Ventures’ return for the year ending June 2024 lists $4,533,333 paid to Pattern Labs Tech Inc. and another $2,266,667 approved for future payment. Its next return records the second payment. The expenditure-responsibility statement supplies both dates: 5 March 2024 and 4 February 2025. Together, they account for the previously announced $6.8 million award.
The later statement describes a grant for software tools, products and analysis concerning global security. It also records $2,639,613 expended by the grantee, a report dated 15 March 2025, and no reported diversion of funds. Both returns repeat that expenditure figure and report date. They do not provide a current balance.
This gives us something much more useful than a grant listing: a named reporting relationship and a specific report to pursue. Under the IRS expenditure-responsibility framework, the foundation is expected to obtain a written commitment to the grant’s purpose, annual spending and progress reports, and access to relevant records. The foundation must retain the agreement and reports. These are the framework’s requirements; the signed Pattern Labs agreement is still missing. Its particular intellectual-property and commercial-use conditions remain an open question.
£459,000 to help construct the measuring instrument
The newly recovered award is Accessible Cyber Attacks Evaluations and Research, awarded to Pattern Labs on 17 September 2024. Its value is £459,000. It is a separate contract from the similarly priced £459,875 attacks-and-safeguards award already in our map.
In the contract bundle, Schedule 20 asks the supplier to research priority cyber capabilities, design evaluations and justify their coverage and robustness. It calls for a first outline of an ideal evaluation suite after two months, repeated feedback with DSIT, advice on running the tests and review of other DSIT evaluations and products. DSIT also promises access to its Inspect technical infrastructure.
That is participation in defining the yardstick. Government researchers supply threat scenarios; the contractor helps work out what capabilities matter, how to test them and why the tests should count as adequate. Those choices shape what later results can tell a decision-maker.
The agreement places final output approval and publication decisions with the buyer. It requires independent quality assurance and access to the data underlying outputs for the buyer’s own checks. At the same time, Special Term 1 preserves the supplier’s underlying IP and know-how and permits related evaluation products to be developed and sold.
The exchange is concrete: the government purchases specialist methods and evidence; the supplier retains knowledge that can support a commercial business. Our inference is that this arrangement can strengthen a specialist’s position over time. Repeated access, testing experience and reusable tools can make it an increasingly attractive supplier of the next round of work.
A disclosed place in the framework discussion
An October 2024 government disclosure also names Pattern Labs and Open Philanthropy in its response about invitations to the closed Frontier AI Safety Frameworks conference. It identifies UK AISI and the Centre for the Governance of AI as organizers and gives a capacity of 90. The earlier response describes the primary audience as employees of labs making frontier-safety commitments, with relevant researchers also invited.
This documents an invitation context, rather than verified attendance. It is nevertheless another specific connection: the funded evaluator appears in the organizational list around a government-linked discussion of safety frameworks.
The remaining documents would settle different questions
The 15 March 2025 grantee report and the signed Good Ventures agreement would show the reported work and the actual grant restrictions. They are the direct route to answering how philanthropic support interacted with commercial development.
For contract 6492, accepted deliverables, final sign-off, QA reports and supporting data would show what the government actually received and accepted. The detailed supplier tender is commercially redacted and has precedence for deliverables. The bundle references a security annex that is not included. Its notice and order form also differ on the end date, with the form including an optional extension. We have not established exercise of that option or a later instrument covering 2026 work.
These gaps define the next requests. They do not erase the mechanism already visible: philanthropy financed technical capacity; government procured work on the measurement methods; the supplier retained a route to commercial reuse. Influence here sits partly in the construction of the evidence on which later decisions may depend. The next test is whether we can follow a specific output all the way into an actual decision.