OpenAI puts $7.5M into an external alignment fund it doesn't control

What the grant actually buys

On February 19, 2026, OpenAI committed $7.5 million — roughly £5.6 million at the exchange rate it cites — to The Alignment Project, a fund created by the UK AI Security Institute (UK AISI) with grant administration handled by Renaissance Philanthropy.

The money joins a pool that already exceeds £27 million from public, philanthropic, and industry backers. Individual projects are funded at £50,000 to £1 million, and may get optional compute and expert support on top of cash.

The most specific detail is what the grant is not. OpenAI states plainly that its funding "does not create a new program or selection process, nor influence the existing process." It buys more of the already-vetted projects that made it through UK AISI's review in the current round.

Handing the selection process to a government institute

The structural choice here is the story. A frontier lab writing a seven-figure check normally comes with strings: its own RFP, its own reviewers, its own topic priorities. OpenAI explicitly declined all three, routing the money through UK AISI's existing pipeline instead.

OpenAI's stated reason is that UK AISI already has the machinery: a cross-sector coalition across government, academia, philanthropy, and industry, a grantmaking pipeline in motion, and a backlog of proposals that have cleared expert review. As a body inside the UK's Department for Science, Innovation and Technology, it also carries a public mandate for serious AI risks.

For a company frequently accused of steering safety discourse toward its own methods, funding work it cannot select is a way of pre-empting the conflict-of-interest objection. The credibility of the gesture rests on OpenAI keeping its hands off the vetting — which it says it is doing.

Betting on research that isn't about frontier models

The funded portfolio spans computational complexity theory, economic and game theory, cognitive science, and information theory and cryptography. These are notably not the kinds of problems that require access to a frontier model or a large compute budget.

OpenAI draws that line itself. It argues frontier labs hold a comparative advantage only in work that "depends on access to frontier models and significant compute," and concedes that "in many kinds of useful inquiry, labs do not retain a comparative advantage."

The underlying wager is a hedge against its own roadmap. OpenAI writes that progress "may ultimately depend on fundamental breakthroughs that change the shape of the alignment problem," and that it wants research that "would matter even if today's dominant methods turn out not to scale in the way we expect." It is paying for uncorrelated bets in case its internal approach hits a wall.

A division of labor between internal and external safety work

A healthy alignment ecosystem depends on independent teams testing diverse assumptions, developing alternative frameworks, and exploring conceptual, theoretical, and blue-sky ideas that may not align neatly with any one organization's roadmap.Montana Labs

OpenAI is describing a split it intends to keep. Its internal effort stays "tightly integrated with model building and deployment" through iterative deployment; the exploratory, foundational work gets pushed outward to an ecosystem it funds but does not direct.

For teams tracking where alignment research money flows, the concrete implication is that the largest new external pool is being consolidated inside a government research institute, not a lab. If UK AISI becomes the default clearinghouse for industry alignment dollars, its review criteria — not any single company's priorities — increasingly determine which independent work gets funded worldwide.