Earlier this month, Ethereum Foundation researcher Justin Drake told the industry to start planning for “bunker mode.” HODL FM quoted his post on X:
“IMO it is now reasonable to brace for the possibility that ECDSA breaks before qday, in the worst case in months not years.”
What he means by a break is private key recovery in about a week, on hardware like a large GPU cluster.
Before going further, here is what CoinDesk found: no practical attack on Bitcoin or Ethereum wallet keys has been demonstrated, and none turned up in the research it reviewed. So this is a warning about a possibility.
Why AI and wallet cryptography share a headline
Math is what set this off. OpenAI published 722 manuscripts from an unreleased model that was tested on roughly 4,000 research problems. Some come with proofs a computer can check. Others are unverified and may contain errors, according to the company.
Drake sees these results as a reason to reassess assumptions in cryptography. He pointed to advances in integer multiplication and a result concerning the 3SUM conjecture as examples of mathematical progress that could challenge established expectations. Neither result, however, demonstrates a way to recover private keys from elliptic-curve public keys.
His point is that if beliefs that old can break, the ones under cryptography need another look.
How a wallet protects funds
Your wallet never holds coins. Ethereum’s developer documentation puts it this way:
“You never really hold cryptocurrency, you hold private keys – the funds are always on Ethereum’s ledger.”
Whoever controls the private key controls the funds, so the whole security question comes down to who can produce a valid signature.
The keys are linked in one direction. On Bitcoin a private key is a 256-bit number, with its range set by the secp256k1 ECDSA standard, and the public key is a point on an elliptic curve. Ethereum follows the same pattern: ECDSA turns the private key into a public key, then the address is the last 20 bytes of that public key’s Keccak-256 hash with 0x in front.
How much the public sees differs by network. Bitcoin’s developer guide says P2PKH or P2SH hashes are often handed out in place of public keys, so the real key shows up only when the coins are spent. On Ethereum, a standard transaction signature allows the sender's public key to be recovered. Once that public key is known, an attacker would still need to solve the elliptic-curve discrete logarithm problem to recover the private key using a classical attack.
Bitcoin exposure depends on the address type and spending history: public keys are revealed when certain outputs are spent, which renders some funds more exposed to future attacks than others.
Cryptographic hash functions provide a separate layer of protection by producing fixed-length outputs designed to resist attempts to recover an input or find another input with the same hash.
What AI can and cannot do today
It helps to separate two ideas. AI-assisted cryptanalysis means using machine learning to hunt for weaknesses in a cipher, and researchers do this. Cracking a wallet means pulling the private key out of an address that holds money, and nobody has shown that.
A 2025 paper in Cybersecurity gives a sense of where the research stands. Hou, Ren and Chen used a machine learning-aided method against round-reduced DES, up to 8 rounds, and their abstract says it matched classical methods or beat them by a small margin. DES is a symmetric cipher, and they attacked a weakened version of it. Elliptic curves weren’t part of the study.
The money that has actually been lost came through software bugs. According to TRM Labs, an attacker began draining Coldcard hardware wallets on July 30 through a five-year-old firmware bug and took 1816 BTC. Maker Coinkite suspected AI helped find the flaw. In late July, a volunteer group called the Bitcoin Red Team pointed AI models at 390 Bitcoin software projects. In about 27 hours it logged nearly 5,000 possible flaws, 85 of them critical. Neither case involved recovering a key from a public key.
Drake’s worry is a different, theoretical one.
“Curves carry rich structure, with room for fancy tricks like Schoof, Frobenius, pairings,” he wrote. “By contrast, hashes are designed to minimise algebraic structure.”
If a model can learn that structure, it might find a shortcut nobody has.
Quantum computers pose a separate threat. A sufficiently powerful, fault-tolerant quantum computer running Shor's algorithm could undermine the elliptic-curve cryptography used by Bitcoin and Ethereum. Today's machines are not capable of doing this at the required scale, but the timeline remains uncertain. Unlike Drake's scenario, which asks whether AI could help discover a new classical attack, the quantum threat relies on a different computational approach.
What happens if the cryptography has to change
NIST finalized its first three post-quantum standards in 2024. Of the four algorithms it first picked, three use structured lattices and one uses hash functions. It built in choice on purpose, offering more than one algorithm per use case in case one turns out to be vulnerable.
AI muddies that picture.
“There is a good chance that the concrete security of lattices will take serious hits from the next two years of AI math,” Vitalik Buterin wrote on X.
Vitalik sees fewer openings for shortcuts in hash-based designs, though he allowed that they could be attacked too. Drake would drop structured assumptions altogether, whether from curves, lattices, or isogenies.
The Ethereum Foundation has set December 2029 as its target for quantum-resistant cryptography. Drake wrote that
“Those timelines must now be revisited and accelerated in light of mathematical superintelligence.”
NIST notes that getting a standard into everyday systems has historically taken 10 to 20 years.
In the meantime, Drake suggests large holders gradually move funds to addresses whose public keys have never appeared onchain. Vitalik supports cutting key exposure where it’s practical but warns that rushing can backfire:
“I personally have lost more money in botched migrations than I have lost in all hacks combined,” he wrote.
A fresh address hides its key only until the first spend.
What wallet owners can do today
None of this will make a wallet AI-proof, but it does cover the failures that have cost people money. Keep wallet software and firmware updated, since Coldcard users lost funds to a firmware flaw. A hardware wallet is worth considering: Bitcoin’s guide says these devices only sign, which removes many weaknesses of general-purpose operating systems. Read the transaction details on the device screen. Guard the seed phrase too. The root seed is the only thing needed to rebuild every key in an HD wallet, BIP39 encodes it as 12 to 24 words, and anyone who holds those words controls the funds.
Don’t move funds in a panic.
“Don’t rush. While I believe there is cause for action a rushed migration would do more harm than good. Don’t panic either,” Drake wrote on X.
Defenders get to use AI as well. Core Lightning’s developers issued an emergency warning on Aug. 27 after AI-generated bug reports turned up real vulnerabilities in their software. As for whether AI could eventually compromise wallet cryptography: in theory yes, if a model finds a classical shortcut no human has found, and nothing in these sources shows one exists. If it happened, networks would move to other signature schemes and holders would rotate to fresh addresses. Planning for that is reasonable. Abandoning your wallet today isn’t.
Gunel Ismayilova
Disclaimer: All materials on this site are for informational purposes only. None of the material should be interpreted as investment advice. Cryptocurrency investments involve significant risk of loss. Please note that, despite the nature of much of the material created and hosted on this website, HODL FM operates as a media and informational platform, not a provider of financial advisory services. The opinions of authors and other contributors are their own and should not be taken as financial advice. If you require advice, HODL FM strongly recommends contacting a qualified industry professional.
