OpenAI’s recent release of hundreds of machine-generated mathematical results has prompted a senior blockchain researcher to urge the industry toward a cautious defensive posture.
On October 7, 2026, Justin Drake, a researcher associated with the Ethereum Foundation whose work also addresses Bitcoin security assumptions, publicly called on holders and institutions to begin planning what he termed “bunker mode.”
The trigger was OpenAI’s October 6 publication of a large batch of findings produced by an internal frontier model.
Reporting on the release described roughly 722 manuscripts spanning about 372 families of results, covering areas such as number theory, geometry, combinatorics, and theoretical computer science.
OpenAI indicated that the model had been evaluated on thousands of problems, that many proofs were accompanied by machine-checkable formalizations, and that the average result consumed compute comparable to a few hours of advanced reasoning time.
Earlier related work had included challenges to long-standing conjectures and rapid progress on difficult open problems.
Drake interpreted the pace as evidence that mathematical capabilities are advancing faster than many cryptographic timelines assumed.
In a public post, he recommended a controlled migration of assets toward fresh addresses whose public keys remain concealed behind a hash until they are used.
Large and sophisticated holders, he suggested, should move the bulk of funds to addresses that have never signed a transaction.
After any spend, remaining balances should be shifted again to a new unused address, potentially derived from the same seed.
He stressed that the step does not require new wallets or novel cryptography and should be carried out without haste or panic, because a rushed transfer could itself cause losses.
Today I call upon the blockchain industry to calmly begin planning for "bunker mode". My personal recommendation is to set in motion a controlled mass migration of assets to fresh addresses, i.e. addresses whose pubkeys remain hidden behind a hash.
Holders, starting with large…
— Justin Drake (@drakefjustin) October 7, 2026
Drake argued it is now reasonable to prepare for the possibility that the Elliptic Curve Digital Signature Algorithm (ECDSA)—the scheme securing most Bitcoin and Ethereum transactions—could be broken before practical quantum computers arrive.
In his worst-case framing, fast private key recovery from an exposed public key, on the order of a week using available hardware such as a large GPU cluster, might occur in months rather than years.
He noted that elliptic curves contain algebraic structure that could, in principle, admit unexpected classical attacks, unlike hash functions designed to minimize such structure.
OpenAI’s materials did not claim any attack on ECDSA or RSA; the short timeline remains Drake’s conjecture, not a demonstrated break.
He pointed to trackers of already-exposed Bitcoin public keys and singled out major custodians as candidates to harden cold storage first.
Smaller wallets, he added, retain partial cover from the large set of early exposed addresses.
Critical signers such as oracles could rotate keys or combine signatures with hash-based schemes.
Longer term, he favored moving toward cryptography that relies on hashes rather than structured mathematical assumptions.
The episode fits a wider unease among security professionals.
As AI systems are adopted across research, operations, and infrastructure, assumptions about how long current digital platforms and financial services will remain safe are being reexamined ahead of 2027 and beyond.
Exposed keys, long-lived signatures, and reliance on algebraic hardness are receiving fresh scrutiny even where no immediate exploit has been shown. Drake’s call is therefore best read as a precautionary operational suggestion amid accelerating mathematical automation, not as confirmation that existing wallets have already been compromised.