Texas AI Docket

National Science Foundation funds Rice to let AI propose numerical algorithms only where a proof checker certifies them

Research and scienceUnited States National Science FoundationHarrisClosed

The National Science Foundation made a standard grant to William Marsh Rice University in Houston on August 17th, 2026. The award record describes reinforcement learning and large language models proposing randomized numerical linear algebra algorithms. The Lean proof assistant is required to certify correctness and cost before a candidate counts. The record states that the machine methods can explore designs efficiently but may produce opaque or inconsistent results without dependable guarantees. It also describes separating real algorithmic improvement from variation caused by the model itself. The award runs from September 15th, 2026 to August 31st, 2029.

How to take part

The award record is public at the National Science Foundation award page under award number 2616828. No comment window, hearing or application step is stated in the record.

Where to do it

Where

1 county. In Houston-Pasadena-The Woodlands.

Timeline

  1. decided

    Award date on the National Science Foundation award record

  2. Today
  3. effective

    Start date of the award

    10 days out

How this decision moved

One dated line per check, oldest first. A line that says nothing changed means somebody looked and it had not.

  1. 2026-08-23

    Admitted. The award record was read directly from the agency's own award endpoint.

  2. 2026-08-26

    Checked and unchanged. The decision still stands as decided.

  3. 2026-08-29

    Checked and unchanged. The grant to Rice still requires a proof assistant to certify an algorithm's correctness and its cost before a machine proposed candidate counts for anything. The award record is otherwise unaltered.

  4. 2026-09-01

    Rice's proof-checked numerical-algorithm research award remains active in the federal record.

  5. 2026-09-02

    Checked and unchanged. The decision still stands as decided.

  6. 2026-09-05

    The Rice award still limits the algorithms AI may propose to the ones a proof checker certifies. The condition is unchanged.

The evidence

Every fact above rests on one of these. The words are the source's own.

AI methods, including reinforcement learning (RL) and large language models (LLMs), can explore algorithm designs efficiently, but they may produce opaque or inconsistent results without dependable guarantees.
National Science Foundation, Award Abstract 2616828 Primary source, official · api.nsf.gov
Lean will certify deterministic correctness and cost claims, thereby converting AI-generated candidates into reusable mathematical objects rather than opaque algorithmic programs.
National Science Foundation, Award Abstract 2616828 Primary source, official · api.nsf.gov
Because LLM-generated algorithms and proofs vary across prompts, the LAAM will also develop sequentially valid inference to separate genuine algorithmic improvements from model-induced randomness.
National Science Foundation, Award Abstract 2616828 Primary source, official · api.nsf.gov
"id":"2616828","initAmendmentDate":"08/17/2026"
National Science Foundation, Award Record 2616828, award fields Primary source, official · api.nsf.gov
"awardeeName":"William Marsh Rice University"
National Science Foundation, Award Record 2616828, award fields Primary source, official · api.nsf.gov
"fundsObligatedAmt":"150000"
National Science Foundation, Award Record 2616828, award fields Primary source, official · api.nsf.gov

Questions about this decision

Answered from the record itself. Every answer is assembled from stored fields, so an answer the record has no basis for is left out rather than guessed.

What is this decision?

The National Science Foundation made a standard grant to William Marsh Rice University in Houston on August 17th, 2026. The award record describes reinforcement learning and large language models proposing randomized numerical linear algebra algorithms. The Lean proof assistant is required to certify correctness and cost before a candidate counts. The record states that the machine methods can explore designs efficiently but may produce opaque or inconsistent results without dependable guarantees. It also describes separating real algorithmic improvement from variation caused by the model itself. The award runs from September 15th, 2026 to August 31st, 2029.

Who decides it?

United States National Science Foundation decides. The record names the deciding body for every entry it carries.

Can the public take part?

The award record is public at the National Science Foundation award page under award number 2616828. No comment window, hearing or application step is stated in the record.

Where in Texas does it apply?

It covers Harris County.

Has it been decided?

It has been decided. The dates on the item page carry when.

What happens next?

A effective is set for September 15th, in ten days.

When did it start?

The earliest date on its record is August 17th, 2026.

What kind of decision is it?

It is filed under research and science.

What sources back it?

One source backs it. It is primary.

Is it on the ERCOT grid?

Yes. It sits inside the ERCOT interconnection.

When was it last checked?

Every fact on it was last verified against its source on September 5th, 2026.

Cite this

Texas AI Docket, National Science Foundation funds Rice to let AI propose numerical algorithms only where a proof checker certifies them. Tracked since August 17th, 2026. Last verified September 5th, 2026. https://texasaidocket.com/item/tx-2026-0093/. Reuse permitted under CC BY 4.0 with attribution. The same entry is in the docket JSON as item tx-2026-0093.

Beat

Filed under Research and science, with every other decision on that beat.

Last checked 2026-09-05