Comparison

Money Already Flows to Overseeing AI, and Now to Examining AI Science

The major AI labs describe scalable oversight, human judgment that keeps pace with smarter AI, as a central problem (OpenAI, July 2023; Anthropic, 8 March 2023; Google DeepMind, 2 April 2025). OpenAI wrote in 2023 that “humans won’t be able to reliably supervise AI systems much smarter than us,” and in September 2026 said it cannot assume safety progress will keep pace. Money already flows to the oversight of AI: the UK AI Security Institute has been allocated £360 million from 2026-27 to 2029-30 (written answer, 6 November 2026); METR, an independent AI evaluator, received about $71 million in commitments in the six months to August 2026; and Canada and Germany committed up to CAD 300 million to LawZero’s Scientist AI (September 2026). Examining AI-produced scientific results is now funded as well. On 15 October 2026 the Simons Foundation, Anthropic and OpenAI said they “will commit $15 million in combined funding and resources,” including support for workshops and initiatives that help researchers evaluate recent AI-assisted results. On 26 October 2026 OpenAI committed to ensure funding for outside researchers to examine three of its results, and the $17.1 million AI for Math Fund round (6 October 2026) includes grants for checking and understanding AI-generated mathematics.

Image 0

0

1

Updated 2026-11-17

Contributors are:

Tags

Scientific Oversight of AI-Assisted Research

1Cademy