The takeaway
Frontier labs are now shipping formal math results faster than peer review can absorb them, and academic communities are responding with their own release norms: external shepherds, public proofs and funded conferences.
Why it matters for builders
Two shifts matter for builders. First, model releases now collide inside a single news cycle, so anything that depends on a specific benchmark or proof claim needs a verification step before it enters a pipeline. Second, the labs are adopting community-owned release processes with external shepherds and public artifacts - a template that enterprise AI governance will likely copy.
Anthropic Refutes a Core CS Assumption Hours After OpenAI
OpenAI published solutions to three long-standing problems in mathematics, computer science and physics on Monday. Within a day, Anthropic released its own refutation of one of the assumptions those results touch - the Exponential Time Hypothesis (ETH), a load-bearing claim in complexity theory that has shaped how computer scientists think about what is hard to compute for more than two decades.
Both labs now claim their models produced algorithms that solve a specific class of logic puzzles in "sub-exponential" time - faster than ETH says is possible. If the claims hold, one of the field's most durable intuitions is wrong.
What happened
OpenAI's release was structured differently from its previous proof drops. Mathematicians got the raw output on GitHub. Computer scientists were given an external working group that asked for the proofs to be published and for OpenAI to fund a conference. Physicists formed a third group to decide how their result gets released.
Anthropic's mathematician Levent Alpöge argued on X that OpenAI's solution leaned on pre-existing work. Hours later Anthropic published its own ETH refutation in its formal-math repository.
Why the rush matters
The ETH is the practical backbone of the P vs NP question. Solving a Sudoku is slow, checking a finished grid is instant, and ETH is the strongest evidence that this gap is real for a whole family of problems. Cryptography quietly depends on that gap.
That is why the first question MIT professor Ryan Williams asked OpenAI science lead Dan Roberts was whether any crypto system had just been broken. According to TechCrunch, OpenAI checked and said no.
Williams had previously called an earlier OpenAI proof dump a "DDOS attack on mathematicians." He is now organizing his field's response anyway, together with Russell Impagliazzo and Ramamohan Paturi, who first published ETH in 1999. Their open letter asks researchers for feedback and for norms around AI contributions to their work.
"Some parts are, maybe you could say routine, but then yes, there are two or three alien-looking parts," Williams said. "I also think that this is probably not the final answer either."
What it means for builders
Two things are happening at once. First, frontier labs are now shipping formal results fast enough to collide with each other inside a single news cycle - a release cadence that no peer-review pipeline was designed for. Second, the labs are being pushed into community-owned release processes: external shepherds, public proofs, funded conferences.
William hopes that hits the right balance. "AI is forcing people to be a little more ambitious about the kind of work that they do," he said.
The Automation Brief
Read 5 AI stories instead of 50.
The essential moves in AI agents, models, automation and infrastructure — filtered for builders and operators, with the part that actually matters.
No noise. Unsubscribe anytime.
Editorial notes
Stefan Trbojevic
n8n Lab Editorial
27 October 2026
27 October 2026
Sources
AI disclosure: AI assisted with research and drafting. Factual claims are reviewed by an editor.
