Next generation artificial intelligence (AI): explainable AI

EPSRC has opened the explainable AI strand of its Next Generation Artificial Intelligence programme, a 12,050,000 pound call for speculative, high-risk fundamental research that makes AI systems explainable rather than merely accurate. The call names five research directions: output reasoning, where the system articulates why it produced a given output; mechanistic interpretability, which seeks to understand the internal computations of a model; process examination, which audits how a decision was reached; imposed explainability, which builds explanation into architecture by construction rather than recovering it after the fact; and uncertainty quantification, which characterises what a model does not know. The framing is unusually explicit that incremental improvement is not fundable: applications must demonstrate potential for transformative outcomes, and EPSRC uses the phrase speculative and high-risk fundamental research with the potential to deliver high reward. Awards run up to 24 months with a fixed project start of 1 February 2027, and the maximum request is 602,500 pounds at full economic cost with UKRI funding 80 percent. Early and mid-career researchers are particularly welcomed, and joint leadership is permitted with a designated administrative lead. The deadline is 20 October 2026 at 4:00pm UK time, with an information webinar having been held on 8 September 2026.