Deep Dive: Two AI Labs Say They Broke the Exponential Time Hypothesis
Two AI labs now say their models broke a rule that computer scientists long assumed to be true, and very few people outside the field understand what that means. This report explains the claim, and why the market angle is thinner than the headlines suggest.
What Was Announced
On October 26, OpenAI said an internal model had refuted the Exponential Time Hypothesis (ETH), a 1999 conjecture about how fast computers can solve certain logic problems. The company says the model found the disproof on October 16 and that it was checked the next day in Lean, a language that verifies proofs by machine. OpenAI published the write-up and the Lean proof on GitHub, alongside two other results in mathematics and physics.
Hours later, Anthropic posted its own paper, "A 2^o(n) Time Algorithm for 3-SAT," credited to Claude. It says an internal Claude research model found the core algorithm on October 16, the same date OpenAI gives, using about 14 million output tokens and no human input, and that the result has been formalized in Lean 4. Anthropic cautions that the result does not establish a practical speedup for real-world problems.
Neither claim has been peer-reviewed. Russell Impagliazzo and Ramamohan Paturi, two of the original authors of the hypothesis, and Ryan Williams of MIT are organizing an open, independent examination. They wrote that if the claim is correct, it will change an important part of theoretical computer science.
One piece of context matters for readers. Anthropic filed for an IPO on October 26, the same day, with plans to list on the Nasdaq under the ticker ANT. The company whose model is credited with one of these results is also a company seeking investors, which is one more reason to wait for independent confirmation.
What the Hypothesis Says
3-SAT is a logic puzzle. Given a formula with n true-or-false variables, can you set them so that every condition in the formula is satisfied? The best known methods for the worst cases take time that roughly doubles with every added variable, which is called exponential time. ETH, formulated by Impagliazzo and Paturi in 1999, says nothing can do fundamentally better. No algorithm, it claims, solves 3-SAT in 2^o(n) time, meaning a running time that grows more slowly than any exponential in n.
The stakes for researchers are large. A literature search run by Claude found more than 800 papers that prove at least one result only under ETH or a hypothesis that implies it.
What does not change.
P versus NP. ETH is a stronger assumption than P not equal to NP. Refuting it does not show that P equals NP, because the claimed algorithm is still not polynomial. P still probably does not equal NP, and the $1 million prize for settling that question is untouched.
Cryptography. Neither lab's paper mentions cryptography, and modern encryption does not rest on ETH.
Everyday software. Nothing changes for anyone running a SAT solver this week. A subexponential algorithm can still be extremely slow in practice.
What Is Verified and What Is Not
The honest status is that these are claims with unusually strong supporting material, not settled results.
No timetable for the independent review has been announced. Until it reports, the right label is "claimed," even though both labs formalized their proofs.
The Market Angle: IBB and CIBR
What is actually moving markets this week. Anthropic's IPO filing lifted AI stocks on Monday, with Nvidia up almost 3%, and the Nasdaq added another 1.2% on Tuesday. Analysts describe the broader market as positioning for a 10-day window of earnings, central bank decisions and US politics. IBB and CIBR rose with that tech rebound, so their gains cannot be pinned on the ETH news.
AI headlines have cut both ways for cyber. In February, Anthropic's launch of Claude Code Security triggered a cybersecurity selloff, with Qualys down as much as 12% and Okta about 10%. Reports in March of a cyber-capable Anthropic model hit the group again. AI capability news is not automatically good for security stocks.
Where the setup stands. Both ETFs are close to their highs after strong years. That makes this a momentum idea, not a bargain, and it leaves little cushion if the theme cools.
What to Watch and the Risks
What to watch:
The independent review. Any statement from Impagliazzo, Paturi and Williams will matter more than any headline. A confirmed error would end the story, and a confirmation would make it far bigger.
Scrutiny of the papers. Both write-ups and Lean files are public, so outside researchers can test them. Look for comments on how large the hidden constants are, since that decides whether the result matters in practice.
This week's market events. The Fed decides on Wednesday, Microsoft and Alphabet report the same day, Amazon follows on Thursday, and Apple reports on November 2. These will move tech more than any math result.
The Anthropic listing. The IPO is expected next month, possibly the week before Thanksgiving, and it is already moving AI stocks.
Risks:
The proofs may not survive review. Claims from AI systems need human experts to confirm them, and neither paper is peer-reviewed.
Narrative trades fade. An idea that links a theory result to biotech or security demand can lose its momentum quickly, and nothing in either paper supports a direct earnings effect.
Both ETFs are near highs. IBB sits about 2.3% and CIBR about 0.9% below their 52-week highs, so a pullback would hit buyers who arrive late.
AI news has hurt cyber before. Earlier AI announcements triggered selloffs in security stocks, so the reaction can go the wrong way.