Overnight, BTC continued its pullback, slipping below the $83k round number. Recently, a remark by Ethereum co-founder Vitalik Buterin brought the once-distant topic of AI and cryptographic security to the foreground. He warned that if AI makes as much mathematical progress in two years as it did over the past fifty, then not only is elliptic-curve cryptography at risk, but even lattice cryptography — long hailed as the post-quantum hope — could be weakened[1]. The backdrop to that judgment is a batch of AI-generated mathematical results OpenAI published a few days earlier, and an even more aggressive warning issued by Ethereum researcher Justin Drake in response.

OpenAI’s “Math Tsunami”

On October 6, OpenAI announced news that sent shockwaves through the mathematics community: its internal frontier model generated 722 mathematical papers spanning algebra, number theory, and theoretical computer science, totaling 372 groups of results[2]. Some papers shipped with Lean formal verification, but OpenAI conceded that unformalized proofs could still contain errors. These results were produced by an undisclosed model, with each result costing compute roughly equivalent to several hours of ChatGPT Pro “thinking” on average. A month earlier, the same model reportedly mobilized over ten thousand AI agents across 88 hours to tackle the famous Navier–Stokes equations problem[2]. This wave of releases drew strong reactions from the math community — both awe at AI’s capabilities and unease over the direct publication of AI-generated results without peer review.

Justin Drake’s “Bunker Mode” Warning

Ethereum researcher Justin Drake quickly tied this event to crypto security. On X, he called OpenAI’s release evidence that “mathematical superintelligence has already arrived,” and warned that the Elliptic Curve Digital Signature Algorithm (ECDSA) — the cornerstone security primitive behind Bitcoin and Ethereum wallets — could be broken in “months, not years”[3]. His logic: elliptic curves carry rich mathematical structure that can be exploited by tricks like the Schoof algorithm, Frobenius endomorphisms, and pairings; by contrast, hash functions are deliberately designed to minimize algebraic structure and are therefore more resistant to this class of attack[3]. Drake called on the industry to enter a so-called “bunker mode,” advising large holders to migrate assets in batches to fresh addresses that have never been used to sign a transaction, since such addresses have not yet exposed their public keys on-chain and exist only in hashed form[4].

The Cryptographers Push Back

Charles Guillemet, CTO of Ledger and a cryptographer, directly pushed back on the sense of urgency. He acknowledged that avoiding address reuse is sensible for large holders, but argued that a mass migration “would more likely cause losses through operational mistakes than defend against the hypothetical threat”[3]. Guillemet contended that treating a classical-algorithm break within months as a planning baseline amounts to FUD (fear, uncertainty, doubt). He further exposed a logical hole in Drake’s argument: if AI truly undermines our confidence in mathematical intuition, wouldn’t that same undermining also apply to hash functions? Yet hash functions are precisely the security foundation of the P2PKH addresses Drake recommends migrating to. In other words, if the mathematical structure of elliptic curves can no longer be trusted, on what basis should hash functions be exempt? Therefore, migrating to addresses with unexposed public keys does not actually neutralize the very threat Drake worries about, and may instead introduce new risks through hasty action[3].

A Rational View, Cautious Action

To be fair, the warnings from Vitalik Buterin and Justin Drake look more like thought experiments about extreme-risk scenarios than imminent predictions. The mainstream view in cryptography is that a classical-computer break of ECDSA remains extremely unlikely in the short term. Vitalik Buterin himself stressed that assets should not be moved in panic, and admitted that his own losses from botched migrations exceed the total from all hacking incidents combined[1].

For ordinary holders, the simplest and safest approach is to keep assets in addresses that have never sent a transaction. Modern Bitcoin address formats (such as P2PKH addresses starting with “1” or SegWit addresses starting with “bc1q”) hide the public key behind a hash; the public key is only exposed on-chain when the address is spent from[5]. Therefore, as long as an address has never initiated a transaction, its public key stays unexposed and is relatively safer. After receiving funds, simply avoid sending from that address to preserve this protection.

The true value of this debate may not lie in whether AI will actually crack ECDSA within months, but in the reminder it offers: cryptographic security rests on hard mathematical problems, and AI is advancing mathematics itself at an unprecedented pace. Preparation is wise, but panic and hasty action are often more real traps than the hypothetical threat itself.

References

  • [1] Vitalik Buterin’s warning (X, Oct 7, 2026) that AI-driven math could weaken lattice cryptography within two years and raise risks for ECDSA: Crypto Briefing coverage.
  • [2] OpenAI’s October 6, 2026 release of 722 manuscripts / 372 result families from an unreleased frontier model (openai/math on GitHub): Inside AI · Unite.AI.
  • [3] Justin Drake’s “bunker mode” post (X, Oct 7, 2026) and Ledger CTO Charles Guillemet’s rebuttal: Decrypt · Cointelegraph.
  • [4] Drake’s specific guidance on migrating to fresh addresses whose public keys stay hidden behind a hash (same sources as [3]).
  • [5] Bitcoin address formats (P2PKH / SegWit) and how they keep the public key hidden until first spend: Bitcoin Wiki — Address.