Summary

  • FRIDA’s 2026 call offered two thematic categories and grants only; its 2025 call had four categories and both grants and awards.
  • In 2026, applicants were slightly more likely to choose AI than Internet Stability and Security, while the selected-project page lists six projects in the latter category and two in the former. The public record does not explain the difference or the narrower call.

The invitation changed before the portfolio did

FRIDA received 57 applications in 2026. Thirty were submitted under Artificial Intelligence applied to the Internet and Networks; 27 under Internet Stability and Security. The final list looks different: six of the eight selected projects sit in Stability and Security, and two in AI.

That is a visible shift in the composition of the call. It is not, by itself, an explanation of how the jury judged the proposals. Before asking what the selection says about the projects, it is worth asking what FRIDA asked people to propose in the first place.

The previous year’s public application summary described four thematic categories: Internet Stability and Security; Internet Access and Connectivity; Open and Free Internet; and AI applied to Internet and Networks. It reported 184 applications, split between 160 grant applications and 24 award applications. For 2026, FRIDA described a grants-only call with two categories: Stability and Security, and AI applied to the Internet and Networks. Access and Connectivity and Open and Free Internet were not listed as categories in this cycle.

This is a change in the program’s published invitation, not proof that FRIDA or LACNIC stopped supporting access, openness, or digital rights elsewhere. The documents reviewed do not say why the two categories or the award track were absent. Nor can the drop from 184 applications in 2025 to 57 in 2026 be read as a clean measure of demand: applicants were offered a different menu and a different funding instrument.

A grant call is a piece of institutional design. Its categories tell potential applicants which problems fit the fund, and its instruments signal whether the program wants to develop a new project or recognize work already under way. When both change in one year, the application total cannot separate a change in interest from a change in who was invited to apply.

A useful disparity, not a verdict

The 2026 pages permit a simple category comparison. The selected list contains six Stability and Security projects from 27 applications, or about 22 per cent. It contains two AI projects from 30 applications, or about 7 per cent. Across the call, eight of 57 applications were selected, about 14 per cent.

Those rates are arithmetic from FRIDA’s aggregate application and selection pages; FRIDA does not publish them as official success rates. They show that the distribution of selected projects did not mirror the distribution of submissions. They do not show why. The public pages provide no category-level score bands, number of proposals that advanced from the first screen, requested or approved amount by category, or written explanation of the final allocation.

The terms of reference describe a two-stage process. An internal screen first checks proposals against general requirements. Applicants who pass are invited to submit a complete proposal for technical review. The first-order criteria include relevance and applicability, impact and results, and continuity and replicability. Innovation, a gender perspective, and alignment with the Sustainable Development Goals are second-order considerations. A regional jury of external experts guides the annual call.

Those criteria could produce different results across the two fields for many legitimate reasons: the fit of particular proposals, the evidence behind their methods, or the balance of the final portfolio. The public record does not let an outside reader distinguish among these possibilities. A 22 per cent versus 7 per cent result is a reason to request the missing context, not to assign a motive to reviewers.

The applicant profile adds context but not an answer. FRIDA reports that 49.1 per cent of proposing organizations came from academia, 31.6 per cent from civil society, 15.8 per cent from the private sector and 3.5 per cent from the technical community; no government organizations applied. Forty-six per cent applied with partners. The page does not break those figures down by category or by the stage at which applications left the process, so they cannot explain the final split.

Selected is not delivered

The selected-project descriptions are plans. One proposes to test active BGP-advertisement techniques for validating filters; another aims to build a regional observatory for route anomalies and RPKI adoption. The AI category includes federated learning on low-cost routers and a local model for WAN management. They are technically specific, but the selection page describes intended work, not a completed deployment or a measured improvement to regional networks.

That distinction matters because the 2026 terms support projects lasting up to 12 months, with grants of up to USD 20,000 and technical follow-up. The maximum is not an award amount for each project. The public project descriptions do not state each budget or a delivery date. A reader should not turn a promising proposal into an operating tool simply because it appears on a selected list.

FRIDA says supported developments and research results must be made available under an appropriate open-source, free-hardware or open license. It also calls for a publishable open-access final report before project close. These commitments make later verification possible, if the program connects the selected-project page to reports, repositories and dated completion status.

FRIDA has operated since 2004 and is now coordinated and funded exclusively by LACNIC. Its annual call is public and uses an external jury. The program therefore has the ingredients for a clear public account of how a funding cycle changes: what the call invited, how proposals moved through its stages, what was selected, and what was eventually delivered. This year’s pages reveal the first and third parts more clearly than the second and fourth.

The fair reading of the 2026 figures is modest. FRIDA narrowed the published call and selected a portfolio weighted toward Stability and Security, even though AI received slightly more applications. The pages do not establish the reason for either decision. They show why a short transition note and a category-level result table would help applicants understand how the fund’s priorities are being expressed.

Sources