Ethereum Foundation researcher Justin Drake has urged the crypto industry to begin preparing for what he calls “bunker mode,” citing that AI-driven mathematical advances could threaten cryptographic systems sooner than the industry expects.
On October 7, Drake recommended that assets be migrated in an orderly manner to new addresses, with the public keys of these addresses hidden behind hashes. He specifically urged large holders to consider moving funds to addresses that have never signed a transaction, and after each signed transaction, to transfer any remaining funds to a new address.
He warned that the industry should consider the possibility that ECDSA could become breakable "within months, not years," and that such a breakthrough could potentially come from classical computing rather than quantum computers.
His argument is based on recent advances in AI for mathematics. Drake highlights breakthroughs related to integer multiplication, 3SUM, and the Erdős unit distance conjecture, suggesting that AI may have discovered mathematical shortcuts previously overlooked by humans.
He believes elliptic curves are particularly worth examining due to their unique mathematical structure and recommends that the industry accelerate its transition to hash-based cryptography.
Vitalik takes this warning seriously.
Ethereum co-founder Vitalik Buterin opposes the notion that users should immediately rush to transfer their funds, but he agrees that AI-driven mathematical advances warrant serious attention.
Buterin believes the industry should consider AI vulnerabilities alongside quantum vulnerabilities. He also questions the assumption that lattice cryptography will remain secure simply because elliptic curve cryptography is currently under greater scrutiny.
However, the programmer still warned users not to rush the migration, noting that he “lost more money due to a poor migration than I have from all hacker attacks combined.” He later elaborated that multisig users should ideally rotate their signing keys after each operation and continue collecting signatures off-chain whenever possible.
Not everyone considers this an emergency.
The cryptographic community has shown a clear divide in its response. Yehuda Lindell, Coinbase’s Head of Cryptography, argued that Drake’s claims are unfounded, stating there is currently no evidence that the decades-old assumptions underlying elliptic curve cryptography have been compromised.
Lindell believes that AI's ability to prove difficult mathematical theorems does not mean it can solve problems like the elliptic curve discrete logarithm. He also notes that if elliptic curve cryptography were truly broken, the consequences would extend far beyond the encryption industry, affecting banking websites, financial systems, software signing, and other internet infrastructure.
Dankrad Feist, who has long been involved in Ethereum development and research, offered a more direct perspective: he argues that if an attacker can immediately compromise any exposed public key, transferring funds to a protected address may not help, as the underlying asset itself could lose all its value.
Dragonfly managing partner Haseeb Qureshi holds the opposite view, stating that Drake’s warning is a cautious response to rapid advances in mathematics and suggesting the industry adopt a more defensive stance for now.
How many cryptographic attacks might AI have already hidden?
The discussion also turned to a more difficult question: How many new cryptographic attacks may already exist but have not yet been disclosed?
In response to the question of how often such attacks might occur, Matthew Green, Associate Professor of Computer Science at Johns Hopkins University, expressed broader concerns, stating that he believes the industry could "lose public-key cryptography," particularly public-key encryption.
Solana co-founder Anatoly Yakovenko countered this premise, arguing that AI is unlikely to simply solve discrete logarithms, integer factorization, or the P ≠ NP conjecture, and is more likely to uncover vulnerabilities in specific curve parameters or groups.
This distinction is at the heart of the debate. AI may identify weaknesses in the implementation of cryptographic systems without overturning the underlying mathematical assumptions.
Solana states that its design provides a "clear path forward."
Solana Foundation's Vice President of Technology, Jacob Creech, denied that Solana needs to adopt Justin Drake's proposed "bunker mode."
Creech believes that Solana's use of Ed25519 provides users with alternative recovery paths, even if the underlying elliptic curve is eventually compromised. The signing key is derived via hashing from a secret seed, and the seed itself never appears on the chain.
According to Creech, the network upgrade would allow legitimate owners to prove ownership of their seed via a hash-based proof and migrate their accounts to a more secure signing scheme. If Solana’s cryptography were somehow compromised under extremely unlikely circumstances, it would be the validators—not users—responsible for upgrading the infrastructure. Creech also noted that “a working proof of concept already exists,” pointing to Anza’s Proof of Knowledge Seed protocol.
Helius CEO Mert Mumtaz also urged users not to follow Drake’s advice, noting that the immediate risks of hasty fund transfers may outweigh the potential gains from a cryptographic breakthrough overnight.
Is this an encryption issue or an internet issue?
This debate ultimately extended far beyond Bitcoin, Ethereum, or Solana themselves. If AI truly finds an efficient way to break widely used public-key cryptography, the consequences would ripple through banks, brokerage platforms, payment networks, TLS, software updates, and other critical internet infrastructure.
However, centralized companies and institutions may respond more quickly to any threats. Blockchains face greater challenges because decentralized networks require coordination and consensus to implement fundamental changes. This makes it especially important to have contingency plans ready before such scenarios arise.
Solana already has a solution. Its Ed25519 design allows users to prove control over the underlying seed and migrate to a more secure signing scheme via network upgrades, with a ready-to-use proof of concept available.
There is currently no public evidence that AI has broken ECDSA or other major cryptographic hardness assumptions. Solana co-founder Anatoly Yakovenko described this panic in a recent tweet as "fear stemming from the uncertainty of smart people not being able to predict the future," but he still urged developers to implement post-quantum cryptography solutions.
Currently, the focus of the debate is on how seriously the industry should prepare for a possibility that researchers cannot rule out but also cannot prove.