Get Up to $100,000 for AI Safety and Science: Foresight Institute Nodes Grant 2026 ($3M Available)

Captured cycle: This page is retained for historical guidance. Confirm whether the program has reopened before planning an application.

Get Up to $100,000 for AI Safety and Science: Foresight Institute Nodes Grant 2026 ($3M Available)

Overview

The Foresight Institute runs an ongoing AI for Science and Safety Nodes program in 2026 that is explicitly different from a standard single-purpose grant round. It is built as a node-based support system, not just a funding call. The page describes three resources together: grants, office/community space, and local compute. The funding is tied to the same ecosystem, and the program is designed around researchers who can benefit from that physical and community infrastructure.

The opportunity is aimed at people who are advancing AI safety, security, and science-focused projects with AI as the core method, not just a peripheral tool. The page repeatedly emphasizes that the “AI-first” principle is important across applications, and that peer collaboration in hubs is expected to be part of value creation.

Before deciding to apply, treat this as a multi-dimensional program: you’re not only applying for money, you are applying into a working community model with monthly cycles, shared activity, and local expectations.

At-a-Glance

What this program actually funds (and what is distinctive about it)

Most grants are straightforward: applicants submit a proposal and receive money if selected. This one is not only that. The Foresight page says the Node model combines:

• grant support,
• office/community space,
• and local compute.

That structure changes how you should think about a proposal. If your work already assumes a team is spread across time zones with limited community engagement, your fit may be weaker than you expect.

The program also explicitly says that grantees are expected to engage with the node environment in Berlin or San Francisco, and to participate in travel-paid workshops there. The page states that funding-only applications are only accepted in exceptional situations. So the project’s real “unit of value” is usually a mix of funding plus collaborative leverage.

Who should apply

The strongest applicants typically look like one of the following:

• a small team with a clear short-to-medium term build plan,
• a solo researcher with a concrete execution plan and realistic milestones,
• a group that can use compute and community support to run experiments quickly.

The opportunity appears broad by domain but narrow in orientation. “AI-for-safety” does not mean any AI work qualifies automatically. The program emphasizes AI-first work and explicit progress in one of its focus areas. If AI is only used for routine data processing and the scientific question itself is unchanged from non-AI variants, your application may be judged as a weaker fit.

A practical way to test if you are a target applicant:

• Can you explain what your project is doing that would not exist, or would be materially delayed, without advanced AI?
• Can you state why node participation changes your probability of success?
• Can your team show measurable deliverables in the next 1-3 years?

If the answer is “yes” to all three, you are likely in the right league.

Who the program says can apply

From the official FAQ, the program states:

• Individuals, teams, and organizations can apply.
• Both non-profit and for-profit applicants are accepted.
• For-profit applicants should justify why they need grant funding.

That means there is no automatic exclusion for startups or commercial entities, which is important if your work is in an applied pathway. What matters is fit and rationale, not legal structure.

Focus areas: practical interpretation

The program lists seven focus areas and uses them as a real filter for review. Use these as your first framing step rather than trying to write your idea first and fitting it later.

1) AI for Security

The program mentions self-improving defense systems, automated vulnerability discovery, formal proofs, and red-teaming workflows. Good fit examples are AI-driven security tooling that closes gaps at speed and scale.

2) Private AI

Projects in privacy-preserving AI, confidential compute, and trust-distributed infrastructure belong here, especially if they improve secure AI operations.

3) Decentralized and Cooperative AI

If your work creates AI systems that coordinate, negotiate, or align around human goals across distributed actors, this area is a strong match.

4) AI for Science & Epistemics

This area includes core science workflow acceleration: data systems, tooling, forecasting, and epistemic infrastructure. It is often a better match for tool builders than narrow application-only projects.

5) AI for Neuro, BCI, and Whole Brain Emulation

Projects in biological intelligence modeling, brain-computer interfaces, and related simulation systems are included, especially where AI makes measurable progress over current methods.

6) AI for Longevity Biotechnology

AI applied to longevity biotech is supported, with smaller amounts mentioned relative to some safety areas.

7) AI for Molecular Nanotechnology

Projects in AI-guided design and simulation/assembly of molecular systems are included, again listed as comparatively smaller grant-size priorities.

What you get from the nodes

Direct program supports

The program explicitly says grants are supported by

• local office space,
• local compute support,
• and community/programming in the nodes.

The compute support is not a generic promise; it is tied to declared need and purpose. You are expected to specify compute requirements and explain how they map to milestones.

Indirect program supports

The value is also in events and cross-project collaboration. The page says grantees are expected to join one travel-paid workshop and can propose additional sprints or collaborative sessions. You should treat that as an explicit time commitment, not an add-on.

Funding terms

The page states:

• most grants are paid as one lump sum,
• for larger multi-year projects, payments may be milestone-tranches,
• later tranche payments are tied to milestone completion and reporting.

The program also says overhead can be covered up to 10% of direct research costs if directly linked to project execution.

Application process (confirmed details only)

Officially stated process

The page gives a clear baseline:

1. Apply through the Airtable form linked on the page.
2. Deadlines are at the end of each month.
3. Applications are reviewed monthly until node capacity is reached.
4. Approximate review time is about two months after a deadline.
5. A fast-track option can be requested, but the program may not honor it.
6. Smaller funding requests can review faster.

Importantly, the page states that they are unable to provide individual feedback to unsuccessful applicants. That affects your planning: do not wait for extensive rejection coaching before improving your next submission.

Review flow (what matters in practice)

From the page:

• First, in-house fit and quality screening.
• Then, potential technical advisor review.
• Potential follow-up: written questions or short call for advancing proposals.

This means the first pass must be legible and internally complete. If the in-house reviewer cannot immediately understand your project, milestones, and need, it is unlikely to advance far enough for a technical review.

Due diligence and compliance requirements

Successful applicants must pass due diligence. The site lists examples of required disclosures and documentation, including:

• confirmation of relevant connections to Foresight,
• disclosure of ongoing legal proceedings such as criminal issues or bankruptcy,
• an itemized budget,
• a project plan,
• organizational documents.

You should not treat this as bureaucratic filler. If your proposal is promising technically but weak on these basics, it can fail later in the process.

How to decide whether it is worth your time

Use this before submitting to avoid wasted cycles.

Strong indicator of good use-of-time

• You can describe a concrete plan with measurable milestones in 1–3 years.
• Your use case is AI-first.
• You can explain how local peer collaboration would materially improve execution.
• You have a realistic budget with direct costs mapped to activities.

Weak indicator of poor fit

• You are applying mostly for unrestricted money.
• You are not clear on where node participation improves outcomes.
• You have no short-term milestones or deliverables.
• The core value proposition can be written in broad claims without evidence of implementation.

A practical fit test

Ask yourself:

• If selected, what changes in the next 60 days?
• What specific outputs would you deliver by the first node deadline cycle?
• What resources (compute, workspace, mentorship, collaborator access) are unblocked by joining a node?
• What does failure look like, and how will you pivot?

If you cannot answer these with specifics, it may be better to refine your idea before applying.

Common mistakes (and how to avoid them)

1) Treating AI as a label instead of a method

This program repeatedly prioritizes AI-first projects. Applications that treat AI as incidental often read as weak fit. Write your proposal so that the proposal’s method depends on AI capability.

2) Ignoring the monthly rhythm

Deadlines happen monthly, and review can take around two months. If your project plan is written as a single annual arc with no monthly milestones, you may not align with the expected operating style.

3) Under-defining milestones

The program’s review criteria include feasibility and milestone quality. Avoid vague target statements; use clear checkpoints and success conditions.

4) Applying for compute without purpose

Because compute is part of value, applicants must specify amount, duration, and purpose. A broad “need compute” request without a workload model usually weakens credibility.

5) Not justifying for-profit need for grants

For-profit applicants are allowed but expected to explain why grant funding is necessary. If your model can generate private financing without this grant and no public-benefit angle exists, this may reduce fit.

6) Omitting reporting obligations

The page says grantees are expected to provide progress updates and tie them to milestones. If your operational plan does not include reporting rhythm, revise before submission.

Preparation checklist before hitting submit

Build a clean packet that maps directly to the official process.

1. One-page objective and problem statement.
2. One to three clear technical approaches and why AI is central.
3. Milestone map with target dates tied to monthly cycles where possible.
4. Quantified budget (direct costs + up to 10% allowable overhead where justified).
5. Compute plan with expected tasks, expected hours/throughput, and privacy/security assumptions.
6. Node engagement plan (Berlin, San Francisco, or hybrid participation model).
7. Risk section: what could go wrong in 30, 60, and 120 day windows.
8. Exit criteria: what success or failure means at each stage.
9. Draft response to evaluation criteria:
   • impact on AI risk reduction (or corresponding science impact for non-safety tracks),
   • feasibility in near term, typically 1-3 years,
   • alignment with one of the focus areas,
   • ability to execute,
   • high-risk/high-reward potential,
   • open source preference where possible.
10. Organization details needed for due diligence (where appropriate): legal status, governance docs, and disclosures.

Suggested application flow after submission

Because direct feedback is not expected for unsuccessful applications, treat your submission as part of a cycle:

• Submit when your proposal is first coherent with milestones and budget.
• Track the review date and do not overfit to one timeline.
• If no award, use the feedback from final decisions (internal notes, rejection reasons if any) to produce the next version.
• Reapply at a later monthly deadline with a revised plan.

The program explicitly says monthly applications continue until capacity is reached, so you do not lose everything by missing the current monthly window unless you are far behind in execution readiness.

FAQ (facts from the official page)

How much funding can be requested?

Approximately $3M is awarded annually. Typical grant levels are between $10,000 and $100,000, with higher amounts more common in AI safety-focused areas.

Is there a final or one-time application deadline?

The page states applications are on the last day of each month. The listed “next application deadline” is May 31.

Can I apply only for money and skip node participation?

The page says funding-only applications are accepted only in exceptional cases. Active participation in SF or Berlin is strongly preferred.

Do they fund compute and office space?

Yes, these are explicitly part of the node model. Eligible projects can request compute budgets or access.

Can for-profit organizations apply?

Yes, but they must justify why they need grant funding.

What are review criteria?

Impact on reducing AI risk (where relevant), feasibility within short timelines, alignment with focus areas, execution capability, high-risk/high-reward profile, and preference for open source unless there is a reason not to.

What happens after selection?

Grantees submit progress updates, and payments can be single or tranche-based depending on size and duration. Overhead up to 10% of direct research costs may be covered.

What is still unclear from the public page

Some practical details are not publicly detailed:

• exact review quotas,
• exact number of awards per cycle,
• explicit country or citizenship restrictions,
• detailed scoring weights by criterion,
• template documents for each application component.

If these are important to you, treat any omitted items as unknown in your planning and confirm with the official page’s application form fields before submission.

Suggested next actions

1. Open the official program page and bookmark it.
2. Confirm the current “next deadline” and whether your preferred node engagement is feasible.
3. Build the proposal around one or two focus areas instead of many unrelated objectives.
4. Draft a version that includes exact milestones and compute assumptions.
5. Submit through the official Airtable form.
6. Note the two-month review expectation so you plan follow-up tasks in the interim.

Official links

• Official program page: https://foresight.org/grants/grants-ai-for-science-safety/
• Application form: https://airtable.com/appyVXc5SMPAvIKpP/pagzBRWeiG3HjH6Qn/form

If you want the most current status, recheck both links before submission and use the latest form fields shown on the Airtable page. The page provides the last confirmed public guidance but can evolve as cycles progress.