OpenAI and Anthropic say their AI models have disproved a long-standing hypothesis about the limits of computation. Three computer scientists whose work is tied to the idea told AFP what the claim could mean, and why the research community still needs to assess it.
Watch Desk analysis
What happened
The exponential time hypothesis, or ETH, holds that some problems are so difficult that computers can solve them only by trying nearly every possible answer. Hundreds of studies in computer science and biology rely on it, according to France 24’s account of the AFP interview.
Researchers Russell Impagliazzo and Ramamohan Paturi proposed ETH about 25 years ago. Ryan Williams, who has publicly questioned it, and the two other researchers are helping organise an independent review of the companies’ claims. The report does not provide the proposed proofs or say that the review has confirmed them.
Why it matters
If the claims hold up, they could affect a theory used to frame what computers can feasibly solve. Williams told AFP the result could open the way to more efficient computing across a wide range of problems. Impagliazzo cautioned that most people would see no immediate effect, though later advances might improve data-analysis tools in science.
The question is not just whether a model can produce a striking result. Paturi told AFP that a company’s announcement cannot substitute for the community’s assessment, and that publishing a proof does not give everyone equal access to the tools behind it.
Our read
This is potentially a substantial result, but the word “disproved” is doing work that the independent review has yet to do. The useful takeaway is both the scale of the claim and the fact that experienced researchers are now organising scrutiny. Until that assessment arrives, ETH is under challenge, not safely filed away as settled history.
What to watch
- What the independent review concludes about the companies’ claims.
- Whether the proofs and methods are made available for scrutiny.
- Whether other researchers reproduce the results or develop them further.
Discussion spark: Should an AI-generated proof count as a scientific breakthrough before researchers can reproduce and explain how it was found?
Sources and evidence
- Shock, vindication, regret: Scientists react to AI roiling life's work (30 October 2026, 05:55 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.