Search and apply for every grant in existence.

This is the one technical area of ARIA's Safeguarded AI programme that remains open - TA1.1, TA1.2, TA1.3, TA1.4, TA2 and TA3 are all archived or cancelled - and it is worth understanding what distinguishes it from the general run of AI safety funding. The work is not empirical alignment research or evaluations. It is building production-grade, security-critical software components whose key security properties are backed by machine-checked proofs, with AI used to assist the formal verification itself. That places it at the intersection of formal methods, systems security and AI, and a team without genuine proof-assistant capability will not be competitive regardless of its AI credentials. The structure is unusual and should be priced in before applying. Work runs in 8-week sprint cycles, and blue teams building the secure components are stress-tested by a funded red team doing adversarial penetration testing. This is an adversarial, milestone-driven programme rather than a research grant that pays out and waits for a paper. The solicitation is rolling, assessed in quarterly batches, with the documented final deadline 31 October 2026 at 14:00 GMT - so applying early is materially better than applying at the deadline, since capacity fills as batches are decided. Eligibility is genuinely international: single organisations or teams spanning academia, industry and non-profits may apply, and there is no requirement to be UK-based, which distinguishes this from many UK government research schemes. Applications go through ARIA's grant platform at aria.grantplatform.com. Anyone considering this should read the full technical area description rather than relying on the summary, because the boundary between what counts as machine-checked and what counts as AI-assisted-but-unverified is exactly where proposals are won and lost.

EQUAL Compute Network is an IDRC-funded initiative to address compute inequities limiting Global South participation in frontier AI research. The network supports researchers, public-interest compute providers, and policy actors in Africa, Latin America, South Asia, and Southeast Asia to evaluate compute needs, build shared compute infrastructure, develop pooled access models, and inform national AI compute strategies. Sub-grants fund research nodes, pilot deployments of shared compute clusters, and policy analysis on compute access. Aligns with the wider AI4D portfolio and complements Canadian and EU compute sovereignty initiatives.

The Bias Bounty Mapping Equity Challenge, hosted on Zindi with Humane Intelligence and funded by the Heising-Simons Foundation, asks data scientists to estimate coverage gaps in Overture Maps across US Census tracts. The aim is to test whether communities most exposed to climate risk are also the least well mapped. Participants predict a composite gap score from 0 to 1 per tract, covering roads, building footprints and points of interest, across four regions: Maricopa County, the Northern California fire corridor, Eastern Oklahoma tribal areas and South-Central Texas. Submissions are scored by RMSE and also receive an automated equity scorecard across urban/rural, tribal and social and climate vulnerability strata. The US$10,000 prize pool pays $4,500, $2,500 and $1,500 to the top three, plus $1,000 for Best Bias Discovery and $500 for Best Documentation. Only open-source tools are allowed, and teams can have up to four people. The challenge runs from 28 August to 31 October 2026. It suits data scientists working on algorithmic auditing, equitable geospatial data and civic AI.