Editorial Note: This article examines the October 2026 preprints and formal Lean 4 repositories released by OpenAI and Anthropic alongside academic commentary from theoretical computer scientists as of October 30, 2026. Formal peer review of the underlying complexity bounds remains ongoing.
Quick Answer (Verified October 30, 2026): Searches for “wired an ai breakthrough” surged following a viral late-October 2026 report by WIRED titled “An AI Breakthrough Has Made Decades of Computer Science Research ‘Irrelevant'”. The story covers a historic announcement on October 26, 2026, in which both OpenAI and Anthropic independently reported that their frontier AI models had discovered and formally verified a mathematical proof refuting the Exponential Time Hypothesis (ETH)—a 27-year-old foundational assumption in theoretical computer science. While the discovery is a landmark moment for AI-assisted mathematics and formal verification in Lean 4, complexity theorists emphasize two critical facts: it does not prove $P = NP$, and it does not break modern encryption or speed up everyday software because the theoretical speedup only kicks in at astronomical problem sizes (beyond $10^{108}$ variables).
1. What Is the “WIRED: An AI Breakthrough” Story About?
In late October 2026, WIRED published an article chronicling the shockwave that hit academic computer science departments after OpenAI and Anthropic simultaneously unveiled preprints claiming to disprove the Exponential Time Hypothesis (ETH):
- OpenAI’s Preprint (October 26, 2026): Titled “The Exponential Time Hypothesis Is False,” generated autonomously by an internal reasoning model on October 16, 2026, and shared for review with complexity pioneers Russell Impagliazzo, Ramamohan Paturi, and MIT professor Ryan Williams.
- Anthropic’s Concurrent Preprint (October 26, 2026): Titled “A $2^{o(n)}$ Time Algorithm for 3-SAT,” generated by an internal Claude research model over roughly 14 million output tokens and published alongside an open-source, machine-checkable proof repository in the Lean 4 theorem prover on GitHub.
2. Plain-English Computer Science: What Is 3-SAT and the Exponential Time Hypothesis (ETH)?
To understand why computer scientists were stunned without getting lost in academic notation, you only need to understand two concepts:
What Is 3-SAT (Boolean Satisfiability)?
Imagine you are organizing a massive wedding with $n$ guests and thousands of strict seating rules, where each rule involves 3 guests (for example: “Either Alice sits at Table 1, OR Bob does NOT sit at Table 1, OR Charlie sits at Table 1”). Finding whether there is at least one valid assignment of True/False choices that satisfies every single 3-part rule at the same time is called 3-SAT. In computer science, 3-SAT is the canonical NP-complete problem—meaning thousands of scheduling, logistics, circuit-design, and optimization puzzles can be translated directly into a 3-SAT formula.
What Did the Exponential Time Hypothesis (ETH) Claim?
Formulated in 1999 by computer scientists Russell Impagliazzo and Ramamohan Paturi, the Exponential Time Hypothesis (ETH) conjectured that there is no “shortcut” algorithm for worst-case 3-SAT that avoids exponential growth. Specifically, ETH stated that any deterministic algorithm solving 3-SAT with $n$ variables must take at least $2^{c cdot n}$ steps (for some constant $c > 0$). If you add more variables, the required computation time doubles (or grows by a fixed exponential base) relentlessly.
3. What OpenAI and Anthropic Announced (And Why Both Used Lean 4)
Both OpenAI’s and Anthropic’s AI models constructed a deterministic algorithm that solves (and counts solutions for) 3-SAT in sub-exponential time—specifically bounded by $2^{o(n)}$ (growing more slowly than $2^{c cdot n}$ for any fixed constant $c > 0$ as $n$ approaches infinity). Because ETH states that no $2^{o(n)}$ algorithm can exist for 3-SAT, a valid $2^{o(n)}$ algorithm mathematically refutes both the Exponential Time Hypothesis (ETH) and the Strong Exponential Time Hypothesis (SETH).
Why is everyone talking about Lean 4? In the past, when a human mathematician claimed a major proof, it could take peer reviewers years to read hundreds of pages of dense prose and spot a subtle flaw. Even worse, large language models are notorious for “hallucinating” plausible-sounding math that contains hidden logical leaps. To solve this, the proofs were formalized in Lean 4—an open-source functional programming language and interactive theorem prover originally developed at Microsoft Research. In Lean 4, every single mathematical deduction must compile against strict axiomatic type-checking rules. If a single step is invalid, the code refuses to compile.
For a closer look at how OpenAI and Anthropic models compare in everyday software engineering and technical reasoning workflows, see our hands-on comparison: ChatGPT vs. Claude for Coding and Technical Writing.
4. Why Did WIRED Say Decades of Computer Science Research Became “Irrelevant”?
Why did the *WIRED* headline use the word “Irrelevant”? For the past 27 years, theoretical computer scientists used ETH as a “load-bearing wall” for conditional proofs. Hundreds of published academic papers follow the structure:
“Assuming the Exponential Time Hypothesis (ETH) is true, then Problem X cannot be solved faster than time Y.”
If ETH is false, any theorem that relied purely on “Assuming ETH is true…” loses its foundational premise! However, as complexity theorists quickly pointed out, those decades of research are not actually “trash”—most of those papers proved reductions (mathematical bridges showing that Problem A is at least as hard as Problem B), and those bridges remain mathematically valid and useful even if the underlying baseline shifts.
5. Does This Prove P = NP or Break Modern Encryption? (No—Here Is Why)
Whenever a major complexity theory headline goes viral, readers immediately ask whether the $1 million Millennium Prize problem ($P$ vs. $NP$) has been solved or whether bank encryption is broken. Here is a clear fact-check table:
| Common Viral Claim | Verified Technical Reality (October 2026) |
|---|---|
| “Did AI just prove $P = NP$?” | No. Proving $P = NP$ requires a polynomial-time algorithm ($n^k$). A sub-exponential algorithm ($2^{o(n)}$, such as $2^{O(n log log n / log n)}$) is still vastly slower than polynomial time. |
| “Will this new algorithm make computers or AI millions of times faster tomorrow?” | No. It is a purely asymptotic (“galactic”) algorithm. The mathematical crossover point where this new algorithm beats existing SAT solvers only occurs when formulas exceed roughly $10^{108}$ variables—far more than the number of atoms in the observable universe ($approx 10^{80}$). |
| “Does this break Bitcoin, AES-256, WireGuard, or Passkeys?” | No. First, ETH is a worst-case assumption about 3-SAT, whereas modern cryptography relies on average-case hardness (such as lattice problems, discrete logarithms, or SHA-256/AES block ciphers). Second, an algorithm that only beats brute force above $10^{108}$ variables cannot break 256-bit cryptographic keys. |
Your everyday cryptographic protections—including FIDO2 passkeys on Windows 11, BitLocker AES-XTS disk encryption, and WireGuard and OpenVPN cryptographic tunnels—remain completely unaffected.
6. Frequently Asked Questions (FAQ)
What is the “WIRED: An AI Breakthrough” article about?
The late-October 2026 WIRED article (*”An AI Breakthrough Has Made Decades of Computer Science Research ‘Irrelevant'”*) reports on OpenAI and Anthropic independently announcing that their frontier AI models generated formal proofs disproving the Exponential Time Hypothesis (ETH), a 27-year-old conjecture in computational complexity theory.
What is the Exponential Time Hypothesis (ETH)?
Proposed in 1999 by Russell Impagliazzo and Ramamohan Paturi, the Exponential Time Hypothesis states that the 3-SAT Boolean satisfiability problem cannot be solved in sub-exponential time ($2^{o(n)}$) in the worst case—meaning any general deterministic algorithm must take at least $2^{c cdot n}$ time for some constant $c > 0$.
Does disproving the Exponential Time Hypothesis mean P = NP?
No. Even if 3-SAT can be solved in sub-exponential time ($2^{o(n)}$), that is still much slower than polynomial time ($n^k$). Therefore, disproving ETH does not prove that $P = NP$, though complexity theorists must still verify that the Lean 4 problem definitions match standard complexity assumptions.
Why doesn’t the new 3-SAT algorithm speed up real-world software?
Because the algorithm is what computer scientists call a “galactic algorithm”—its sub-exponential advantage is purely asymptotic and only becomes faster than existing brute-force or heuristic SAT solvers when a problem has more than roughly $10^{108}$ variables, which is larger than any physical computer could ever store.
