← Latest · Archive

SEV-3OpenAI
2 sources standard

OpenAI announced a new initiative to fund independent research on AI alignment, committing $10 million over three years to academic institutions and non-profit organizations. The program will support work on technical approaches to ensuring AI systems behave as intended, including interpretability, robustness, and value alignment research.

The company stated it will prioritize proposals that address "scalable oversight" — methods for evaluating AI systems whose capabilities exceed human ability to directly assess outputs. Funding will be distributed through competitive grants ranging from $100,000 to $1 million per project, with applications opening in March 2026.

OpenAI's announcement follows recent public criticism of the company's resource allocation to safety research. In January 2026, former employees published an open letter claiming internal alignment teams had been "systematically deprioritized" since 2024. OpenAI disputed those characterizations but declined to disclose current headcount figures for safety-focused roles.

The initiative will be administered through a newly formed Independent Research Council, chaired by Professor Stuart Russell of UC Berkeley. Council members will review proposals and allocate funds without OpenAI involvement in individual grant decisions, according to the announcement. OpenAI will retain no intellectual property rights over funded research, which must be published openly.

The company stated the program aims to "accelerate progress on the hardest problems in alignment" and acknowledged that "no single organization can solve these challenges alone." It did not specify whether the $10 million represents new funding or reallocation of existing research budgets.

Applications will be accepted from researchers worldwide, with no restrictions on institutional affiliation. The first cohort of grants is expected to be announced in June 2026.

[source]

Why this is an AI incident

Launch-archive bulk classification (10 May 2026). Source signal originates from a real AI provider, regulator, or model-comparison probe; the harm or behavioural change described would not have occurred without the AI system being deployed in the role described. Editor reviewing the archive may amend the rationale per-wire.

Counterfactual "but-for" test per the Editor's Guide.

Codes M1, F10
Providers OpenAI