Lead – In October 2026, OpenAI announced that its AI systems have produced solutions to three high‑profile mathematical problems, one of which it describes as a “refutation of the Exponential Time Hypothesis (ETH).” Anthropic quickly issued a similar claim, and leading researchers such as MIT professor Ryan Williams have organized a community‑wide response to evaluate the findings, assess their impact on computer‑science theory, and address potential security implications.
AI‑generated proofs and the ETH claim
According to TechCrunch, OpenAI’s release included algorithms that allegedly solve a class of logic puzzles in sub‑exponential time—faster than the exponential time bound stipulated by the ETH. The ETH, first articulated by UC San Diego computer scientists Russell Impagliazzo and Ramamohan Paturi in 1999, has long served as a cornerstone assumption that certain problems require time that grows exponentially with the number of variables.
Anthropic’s mathematician Levent Alpöge, as reported by TechCrunch, suggested that OpenAI’s results may be built on existing work, and Anthropic subsequently published its own “refutation of ETH.” Both companies therefore claim that their models have identified algorithms that breach the previously accepted exponential lower bound.
Structured releases for three disciplines
TechCrunch notes that OpenAI adopted a differentiated distribution strategy for the three fields involved:
- Mathematics – The proofs were posted publicly on GitHub for mathematicians to review.
- Computer Science – A working group was formed, requesting that the proofs be shared openly and that OpenAI fund a dedicated conference to discuss the results.
- Physics – A separate working group was tasked with deciding the appropriate release format for the physics‑related findings.
Academic response led by Ryan Williams
MIT professor Ryan Williams, whose research focuses on computational complexity, is organizing the computer‑science community’s reaction, according to TechCrunch. Williams has previously criticized OpenAI’s “proof dump” approach, calling an earlier October release a “DDOS attack on mathematicians.” Nevertheless, he agreed to collaborate with the frontier labs to help interpret the new solution.
Williams, together with Impagliazzo and Paturi, issued an open letter—again reported by TechCrunch—soliciting feedback from scientists and prompting a discussion of the norms governing AI contributions to scholarly work. In an interview, Williams said, “Some parts [of OpenAI’s proof] are, maybe you could say routine, but then yes, there are two or three alien‑looking parts, and it’s exactly these parts that I saw that made me think, wow, this is something that the whole community needs to work on together.” (According to TechCrunch.)
Context: P vs NP and the significance of ETH
TechCrunch explains that the ETH belongs to a broader family of conjectures surrounding the P vs NP problem, which asks whether every problem whose solution can be verified quickly (NP) can also be solved quickly (P). A familiar illustration is Sudoku: verifying a completed grid is trivial, while finding a solution can be time‑consuming. The prevailing belief, reinforced by the ETH, is that many NP‑complete problems inherently require exponential time for exact solutions, a premise that underpins modern cryptography.
Williams, as quoted by TechCrunch, emphasized that “P = NP is saying that, not only can you verify Sudoku solutions when they exist, you can just generate them… it seems too good to be true.” Consequently, most researchers assume P ≠ NP, and the ETH has provided additional evidence supporting that view.
If OpenAI’s and Anthropic’s claims hold, the ETH would be falsified, potentially opening a pathway toward algorithms that collapse the P vs NP distinction. Williams cautioned, however, that “I do think that it kind of calls into question a lot of the intuitions that we thought we had about … how tractable [hard NP problems] are. I also think that this is probably not the final answer either. I don’t think that OpenAI’s algorithm could possibly be the best algorithm you can get.” (According to TechCrunch.)
Practical implications and next steps
The durability of the ETH means that dozens of papers over the past decade cite it as a foundational assumption. TechCrunch reports that Williams warned many of those works may need to be revisited if the AI‑generated algorithms indeed invalidate the hypothesis. To that end, he has pressed OpenAI to sponsor a conference, arguing that the community must decide “what happens next besides understanding the algorithms. We need to understand, how should the foundations change compared to what we had before?”
Security concerns also emerged. Williams asked OpenAI’s science lead Dan Roberts whether the new algorithms compromise existing cryptographic schemes. According to TechCrunch, OpenAI responded that they had checked for such impacts and were “reasonably confident that this didn’t … imply any breaks.”
Opinion: The evolving role of AI in scientific discovery
This analysis reflects editorial interpretation. The episode illustrates a shifting paradigm in which AI systems can generate results that challenge long‑standing theoretical boundaries. While the claims are extraordinary, the scientific method demands rigorous, peer‑reviewed verification. If validated, the findings could accelerate progress in algorithm design but also compel a reassessment of security protocols that rely on hardness assumptions. Moreover, the episode underscores the need for clear norms governing AI‑assisted research, a topic already highlighted by the open letter referenced by TechCrunch.
Conclusion
OpenAI’s and Anthropic’s announcements have ignited a coordinated response from leading computer scientists, prompting open‑letter calls for feedback, proposals for dedicated conferences, and preliminary security assessments. As the community works to verify the AI‑produced proofs, the potential refutation of the Exponential Time Hypothesis could reshape theoretical computer science and influence the future design of cryptographic systems.
Source: TechCrunch