Debate is intensifying over whether artificial intelligence could threaten cryptocurrency cryptography before quantum computers arrive at scale. According to CoinDesk, Glassnode co-founder Rafael Schultze-Kraft said more than 6 million Bitcoin (BTC) already have fully exposed public keys on-chain, representing about 31.2% of the circulating supply. At the same time, Ethereum Foundation core researcher Justin Drake warned that AI-driven mathematical breakthroughs could challenge traditional cryptographic algorithms in “months, not years.”
Justin Drake calls for “bunker mode”
In a post dated Oct. 7, Drake said AI firms including OpenAI have been producing large volumes of mathematical papers and breakthroughs, citing examples such as advances tied to integer multiplication and the overturning of the 3SUM conjecture. He argued that AI compute and “superintelligence” searching for weaknesses in elliptic-curve cryptography could be far more capable than many expect.
Drake said attackers might be able to combine large GPU clusters with AI-assisted mathematical progress and recover private keys from already exposed public keys in about a week, even before quantum computing becomes widely available. On that basis, he urged large institutions and major holders to move into “bunker mode” and harden defenses.
His recommendations included moving funds to fresh addresses that have never signed a transaction so the public key remains hidden behind a hash, rotating to a new vault after each transaction, having oracles and Layer 2 security councils consider key rotation, and shifting over time toward purely hash-based cryptographic systems.
Vitalik Buterin says the risk is real, but warns against panic
Ethereum co-founder Vitalik Buterin responded that the community should take AI-related cryptographic risk seriously, but he strongly opposed “panic” wallet migration today.
Buterin said, “I have personally lost more funds due to mistakes than due to hackers.”
He also agreed that the issue is not limited to ECDSA. In his view, even defenses once regarded as secure, including lattice-based cryptography, could face renewed pressure from AI over the next two years.
As for practical steps, Buterin said it is a good habit to keep funds in addresses that have never signed transactions when doing so is operationally simple. He added that multisig setups should use off-chain signing where possible and reduce the exposure window, while structured cryptographic schemes should increase parameter sizes. He also said Ethereum will accelerate its move toward pure hash-based signature schemes such as SPHINCS-.
On-chain data puts the scale of exposure into focus
Glassnode’s figures gave the discussion a concrete base. More than 6 million BTC currently have visible public keys on-chain, equal to about 31.2% of Bitcoin’s total circulating supply.
Since May this year, another 222,000 BTC have joined that exposed category, worth about $18.2 billion based on the figures cited in the report. The increase has outpaced newly mined block rewards over the same period.
The report said public-key exposure usually comes from address reuse or from older P2PK and Taproot transaction formats. Under normal assumptions, possession of a public key does not let an attacker move funds. But if AI were to break the underlying cryptographic assumptions, those public addresses would be among the first at risk.
Exposure levels vary sharply across firms
The data also showed large differences across exchanges and institutions.
- Among exchanges, Binance had an exposure ratio of 83%, Robinhood 100%, and Coinbase about 10%.
- Among institutions, only about 2% of the 375,000 BTC managed by Fidelity were exposed, while Grayscale stood at about 49%.
Those figures point to major differences in custody practices and cold-wallet isolation standards.
Not everyone agrees with the warning
The argument has also drawn pushback. Coinbase head of cryptography Yehuda Lindell and others dismissed the claim as FUD, saying the elliptic-curve discrete logarithm problem remains robust and does not support fears that AI could crack ECDSA in the near term.
Even so, the dispute has pushed the crypto community to reexamine how the next generation of security defenses should be built.