Some clarification seems in order, based on reading the thread: the program appears to have hit a hard cap spending limit, so any further spend is frozen until human beings can be brought online to audit the in the unpaid-applicants pool. Existing spend that was issued and accepted by applicants will not be clawed back.

Most likely, the AI hit the brick wall equivalent of ‘your credit card was declined’ and is unable to follow through on issuing credits. I will freely admit my suspicion of an AI running this is in part due to the maximum ironic potential of that guess, but they would never start out using a human for this given their beliefs about AI, and the way things collapsed (approved but unissued) reads like a poorly coded accounting transaction that one would expect from a novice coder with no concept of ‘never overcommit financially’, or an AI. The unstated-as-human audit is an effective CYA as it allows them some latitude in deciding how to proceed, especially if the unpaid applicants pool is 10-100x the size of the budget issued. One wonders what the audit criteria will be: ‘serious’ applicants only? noteworthy applicants, such as well-known projects and the press? and whether they’ll publish any statistic about how many of those audits resulted in approvals. But it seems certain it will, whether it was previously or not, be a human being making those decisions with PR in mind.