UKRI Next Generation AI: Explainable AI 2026: Up to £482,000 for High-Risk AI Research Projects
The UKRI and EPSRC Next Generation AI: Explainable AI grant supports speculative, high-risk fundamental research that could produce a step change in the explainability of future artificial intelligence systems, with applications due 20 October 2026.
UKRI Next Generation AI: Explainable AI 2026: Up to £482,000 for High-Risk AI Research Projects
The UK Research and Innovation (UKRI) and Engineering and Physical Sciences Research Council (EPSRC) Next generation artificial intelligence (AI): explainable AI opportunity is a new research grant for ambitious ideas that could change how future AI systems are understood. The call is open from 11 August 2026 and closes on 20 October 2026 at 4:00pm UK time. Projects must start on 1 February 2027 and can run for up to 24 months.
This is not a conventional software-development competition and it is not a general-purpose artificial intelligence grant. UKRI is looking for speculative, high-risk fundamental research with the possibility of a major advance in explainability. A proposal can explore how a model reaches an output, how to communicate uncertainty, how internal mechanisms produce behaviour, or another technically grounded route that makes future AI systems more understandable. The call is deliberately broad about methods, but it is not broad about ambition: incremental work, routine model improvement, and projects that are primarily social science, legal, policy, or ethics research are outside the intended fit.
Key details
| Detail | Official information |
|---|---|
| Opportunity | Next generation artificial intelligence (AI): explainable AI |
| Funder | UK Research and Innovation and EPSRC |
| Status | Open from 11 August 2026 |
| Application deadline | 20 October 2026 at 4:00pm UK time |
| Project start | 1 February 2027 |
| Maximum project length | 24 months |
| Maximum full economic cost | £602,500 |
| Maximum UKRI contribution | £482,000, equal to 80% of full economic cost |
| Total fund | £12,050,000 |
| Eligible base | A UK research organisation eligible for UKRI funding |
| Application system | UKRI Funding Service; applications cannot be submitted through Je-S |
| Assessment model | Distributed peer review, with applicants also reviewing other proposals |
| Opportunity contact | [email protected] |
The £482,000 figure is the maximum UKRI contribution when an application requests the full £602,500 full economic cost. It is not a guaranteed award amount or a target budget. UKRI welcomes smaller requests, and every cost must be proportionate to the proposed work and justified in the application.
What the grant is designed to support
UKRI describes this call as the first opportunity delivered through its flagship IS8 AI Programme. The programme is intended to respond to strategic and societal needs around increasingly capable AI systems. Explainability matters because a system that cannot be meaningfully understood can be difficult to trust, evaluate, govern, improve, or safely deploy. The official call connects explainability with safety, reliability, responsible use, innovation, public confidence, and future market opportunities.
The grant is especially suitable for a project that could open a new research direction or test a theory that may fail. UKRI explicitly says that supported work is not expected to produce a tangible AI model, spinout, or other final product within two years. It also says applicants should not be penalised for risky ideas that build a pipeline for future impact. That is an important distinction: the proposal should explain the insight the project could establish, not pretend that every ambitious experiment has a predictable commercial endpoint.
The scope includes several illustrative areas:
- output reasoning, including post-hoc explanations, counterfactuals, probabilities, and explanations adapted to different users;
- mechanistic interpretability, examining how internal structures and mechanisms give rise to outputs and behaviours;
- process examination, including how systems use tools or interact with other models and components;
- imposed explainability, such as designing systems to be inherently more interpretable or using physics-informed approaches; and
- uncertainty quantification, including ways to communicate confidence, reliability, and limitations.
These examples are not a closed list. A proposal can take a different route if it makes a convincing technical case for a step change in explainability.
Who this opportunity fits
The lead applicant must be based at a UK research organisation that is eligible for UKRI funding. The organisation must be able to host the project and submit the application through the UKRI Funding Service. UKRI particularly welcomes applications from early- and mid-career researchers, and it encourages researcher co-leads when they have made a substantial intellectual contribution and will remain engaged in the research.
The call allows joint leadership. Where a project uses that arrangement, the applicants must agree which eligible UK organisation will act as the administrative lead and which person will be listed as project lead. The administrative lead organisation is responsible for submission. The project lead may choose to devote a higher proportion of time to the short project than would normally be expected, but that commitment must be justified against the work plan.
International researchers cannot apply as ordinary project leads under this call. They may participate as a project co-lead (international) only through the UKRI-RCN Money Follows Cooperation agreement or the UKRI-IIASA agreement. Other international collaborators, and UK partners that are not based at approved organisations, should be included as project partners where they have an integral role.
There is also a participation limit. A person may be involved in no more than two applications to this opportunity, and only one of those applications may list that person as project lead. Check this with every proposed co-lead and collaborator before finalising the team.
The best fit is a technically strong team that can show why the work is genuinely speculative and why the possible reward justifies the risk. A proposal whose main contribution is an ethical framework, public-attitudes study, legal analysis, policy recommendation, or governance review is unlikely to fit by itself. Interdisciplinary work is welcome, but the project must work toward a demonstrable impact on AI technology and the UK’s capability to design and deploy explainable systems.
Funding, duration, and allowable costs
The full economic cost of a project can be up to £602,500. UKRI will fund 80% of that cost, or up to £482,000, with the standard 20% of full economic cost supplied through the host organisation’s normal arrangements. There is no separate matched-funding requirement beyond that standard contribution. Third-party partners may contribute cash or in-kind support, but reviewers must not treat extra institutional contributions as a reason to score an application more highly.
Projects must start on 1 February 2027 and last no longer than 24 months. The fixed start date makes the preparation of staff contracts, facilities, data access, and partner agreements more important than it would be in a flexible-start competition. UKRI also says that no-cost extensions arising from non-people-related reasons will be allowed under this opportunity, so the delivery plan should be realistic from the outset.
The grant can support contributions to the salaries of the project lead and co-leads, research and technical staff, and project-management staff. It can also cover research consumables, including compute costs; travel and subsistence, including travel associated with the explainable-AI cohort workshops; data preservation, data sharing, dissemination, estates, and indirect costs. Equipment costing more than £25,000 per item, including VAT, is not funded. Smaller equipment should be placed under the appropriate directly incurred cost heading.
The application should not be built around a PhD studentship. UKRI expressly excludes PhD studentship costs. The same caution applies to a budget that is padded with staff, equipment, or compute without a clear connection to the research questions. The cost justification is a test of whether the resources are comprehensive, proportionate, and the best use of the requested funding, not merely an accounting exercise.
How to apply through UKRI
Applications use the new UKRI Funding Service. The host organisation must be registered there; Je-S is not available for this competition. The project lead starts the application, confirms their role, selects the host organisation, and creates or uses a Funding Service account. The application can be saved and worked on offline, but the research office must have enough time to inspect the final version and submit it.
The practical sequence is:
- Confirm that the host organisation is eligible and registered with the Funding Service.
- Agree the project lead, any co-leads, the administrative institution, and the roles of any partners.
- Create the Funding Service application and read all of the question-specific instructions.
- Prepare the Vision and Approach document, team capability evidence, risk-and-reward case, ethics and responsible-research responses, cost justification, reviewer nominations, and partner material as applicable.
- Check the application in the Funding Service read-only view.
- Send it to the research office for institutional review and cost checks.
- Allow the research office to submit the completed application before 4:00pm UK time on 20 October 2026.
UKRI says submitted applications cannot be changed. Internal research-office deadlines are likely to be earlier than the public deadline, so ask for the host institution’s date before committing to the call.
What the application must contain
The central Vision and Approach response must be a single PDF of no more than three sides of A4, single-spaced, in 11-point Arial or an equivalent sans-serif font, with margins of at least 2cm. An additional page may be used for a diagrammatic work plan. The file must be no larger than 8MB and should use the unique Funding Service number followed by “Vision and Approach” as its filename.
The Vision should state the problem, why it matters, what the proposed research could change, and how it would advance explainability in future AI systems. The Approach should explain the framework, methods, milestones, work plan, risks, and reasons for choosing those methods. The proposal needs to be understandable to technically informed reviewers outside the exact niche, while still giving enough detail to show that the work is research rather than a slogan about trustworthy AI.
The Applicant and Team Capability to Deliver section allows up to 1,500 words. It uses the UKRI Résumé for Research and Innovation structure, with up to 1,000 words for the R4RI modules and up to 500 words for additions if needed. Select evidence that demonstrates the team’s ability to deliver this particular project: relevant ideas or methods, constructive collaboration, leadership and development of others, contributions to the wider research community, and work that has benefited research users or society.
The application also includes ethics and responsible research and innovation. Even a theoretical or computational project should consider misuse, security, data protection, reproducibility, and the implications of more capable or more interpretable systems. If the work uses animals, genetically modified organisms, genetic technologies, or human participants, the relevant UKRI sections and supporting information must be completed. If those categories do not apply, the form still requires the appropriate confirmation.
The resources and cost justification has a 1,000-word limit. Explain expensive staff time, unusual consumables or compute, significant travel, facilities, infrastructure, and exceptional costs. The risk-and-reward section has another 1,000 words and is central to this call: describe why the research is exciting, why it is risky, what could constitute a step change, and why the project is more speculative than a normal responsive-mode application.
If project partners are included, each partner must have an integral role. The application records the organisation, contact details, type and monetary value of the contribution, and a letter or email of support. Each support letter is limited to one side of A4 and should confirm the partner’s commitment, explain the value and relevance of the project, and describe the additional contribution. Host organisations and co-leads’ research organisations should not be submitted as partner-support letters.
How the review process works
This call uses distributed peer review rather than a conventional panel stage. Applicants agree to review other proposals submitted to the same opportunity. Each application receives up to eight reviews, and those reviews help rank the proposals for the funding decision. UKRI expects each applicant team to receive approximately eight proposals to assess and provide feedback on.
The reviewing obligation is a real condition of applying. Failing to complete the assigned reviews by the deadline can lead to automatic rejection of the submitted proposal, except in limited extenuating circumstances. Teams must nominate at least one and no more than two people who can participate in the review process, including the project lead and a co-lead where appropriate. Nominees need their name, institution, email address, and the UKRI council area matching their expertise.
Reviewers will examine vision and approach, team capability, ethics and responsible innovation, resources and cost justification, and risk and reward. UKRI will make the final decision and may use portfolio balancing. The distributed process is therefore not a reason to write for one narrow specialist; the case needs to be clear, credible, and assessable across the relevant research community.
A practical preparation strategy
Start by writing the one-sentence research claim: what new understanding of AI explainability could this project produce that current methods cannot? Then separate the claim from the eventual application. The call allows long-term and speculative rewards, so an honest account of what the two-year project will establish is more useful than promising a finished product.
Next, map the risk. Identify the assumption that might fail, the experiment or theoretical test that will challenge it, the information gained if it fails, and the result that would justify a follow-on project. This lets the proposal show that risk is designed and managed rather than simply asserted.
Build the work plan around milestones that fit the fixed 24-month term. Tie each milestone to a method, a decision point, and a plausible output. If the project uses compute, data, or a specialist facility, confirm access before the proposal is written. If a partner is essential, obtain their commitment early and make the partner letter specific. Do not wait for the Funding Service to expose every upload field before contacting collaborators.
Ask the host research office to check the full economic costing and the eligibility of every named organisation. Give the office a near-final draft well before 20 October. Also reserve time for the peer-review commitment: applying creates a second workload in November, and the proposal should not be submitted by a team that cannot meet it.
The opportunity page announces a potential-applicant webinar for 8 September 2026 from 12:30pm to 1:30pm UK time. Researchers with a plausible project should use the webinar and the official guidance to resolve scope or process questions, rather than infer rules from older EPSRC calls.
Common mistakes to avoid
The most serious mistake is presenting incremental work as a high-risk breakthrough. A faster training method, another benchmark comparison, or an ordinary explainability feature may be valuable research, but the application must explain why it could create a new direction or a step change. If the project could be submitted unchanged to a standard responsive-mode scheme, explainable-AI reviewers may question why it belongs in this pilot.
Another mistake is making the proposal mostly about social consequences while leaving the technical contribution vague. The call welcomes interdisciplinary approaches, but it will not fund a majority non-technical project whose main outputs are legal, policy, social-science, or ethics analyses.
Avoid treating the £602,500 full economic cost as an entitlement. Request only what the project needs, justify each unusual cost, and remember that UKRI’s contribution is normally 80%. Do not include PhD studentship costs or equipment above £25,000 per item.
Finally, do not overlook administrative rejection risks: using Je-S, missing the host’s internal deadline, exceeding the three-page Vision and Approach attachment, omitting a reviewer nomination, exceeding the two-application participation limit, submitting uninvited resubmission material, or assuming the application can be edited after submission.
Frequently asked questions
Is this a scholarship or a business grant?
No. It is a UKRI research grant for a team based at an eligible UK research organisation. It is intended for fundamental research and is not a direct award to an individual or a commercial start-up.
How much money can one project receive?
The full economic cost can be up to £602,500. UKRI will fund 80% of that cost, up to £482,000. Smaller proposals are welcome and must be justified; the maximum is not a required request.
Can the project begin later than February 2027?
The official call states that projects must start on 1 February 2027. Treat that as a fixed start date unless UKRI publishes a change.
Can an international researcher lead the application?
Not under the ordinary eligibility route. International researchers may participate as project co-lead (international) only through the named UKRI-RCN Money Follows Cooperation or UKRI-IIASA agreement. Other international collaborators should normally be project partners.
Must a proposal produce a new AI model within two years?
No. UKRI says projects are not expected to produce tangible outcomes such as new AI models or spinouts within the two-year period. The proposal should instead define the foundational insight, test, method, or evidence that the project can realistically deliver.
What happens after submission?
Applicants participate in distributed peer review, usually reviewing around eight other proposals. UKRI then makes the final funding decision and may balance the portfolio. Reviewers and applicants should read the official distributed peer-review rules before submission.
Official links and next steps
Read the official UKRI opportunity page for the live status, Funding Service application link, detailed questions, distributed peer-review rules, and any updates. The named opportunity contact is [email protected]; questions about costings and submission should first go to the host organisation’s research office.
If the project is a genuine technical bet on making future AI systems more understandable, start with the core research claim and a two-year work plan. Confirm the host, team limits, facilities, and review obligations early. Then submit through the UKRI Funding Service before 20 October 2026 at 4:00pm UK time.