As concern grows that artificial intelligence could bring down existing cryptographic systems, a debate is spreading in the cryptocurrency industry over so-called “bunker mode,” a way to protect assets by moving them to new addresses. There is still no evidence AI has actually broken cryptography, and some counter that rushed transfers could instead increase security risks.

Cointelegraph reported on Oct. 14 that Ethereum researcher Justin Drake raised the possibility that AI could find a mathematical breakthrough that neutralises elliptic-curve cryptography even in today’s computing environment. That led to arguments that preemptive steps are needed, such as reducing public-key exposure risks and moving assets to new addresses in a controlled way.

The key issue is whether AI could undermine existing cryptographic systems before quantum computers are commercialised. Major blockchains including Bitcoin and Ethereum use elliptic-curve cryptography to verify transaction signatures and protect ownership of assets. They rely on a mathematical assumption that deriving a private key from a public key is realistically very difficult.

Drake warned that because elliptic-curve cryptography has a more complex mathematical structure than hash functions, AI could discover new ways to attack it. Johns Hopkins University cryptography professor Matthew Green also voiced concern, mentioning the possibility that public-key cryptographic systems could collapse.

Some in the industry counter that no real threat has yet been proven. Coinbase head of cryptography Yehuda Lindell pointed out there is no evidence at all that the mathematical foundations of elliptic-curve cryptography have weakened. Lucien Bourdon, a Bitcoin analyst at Trezor, also said a recent AI-written mathematics paper did not confirm a practical way to attack ECDSA, and assessed Drake’s scenario as assuming a worst-case situation. Bourdon warned there is no need to take special measures right now, and that moving assets in haste could lead to mistakes or fraud losses.

Controversy is also continuing over how effective bunker mode would be. Jack Herbert, co-founder of hardware wallet firm FoundationHQ, said that if a wallet application or a third party has already obtained an extended public key, moving assets to a new address under the same account may not solve the public-key exposure problem. An extended public key is used to generate or verify public keys for multiple addresses derived from a single wallet. If that information has already been exposed, he said, merely changing addresses makes it difficult to completely sever links to existing information. If ECDSA is actually broken, attackers may be able to recover corresponding private keys without stealing the hardware wallet itself. Herbert said the use of new addresses and the extended public key exposure issue can be reviewed in theory, but it is difficult to view this as a practical security measure.

Some also warn that if cryptographic systems are shaken, damage may not be limited to individual wallets. StarkWare Chief Executive Eli Ben-Sasson said simple wallet protection is not enough. Ethereum transaction signatures as well as network infrastructure signature systems, consensus mechanisms and the KZG polynomial commitment method used for blob data could also be affected, he said. The process of replacing or supplementing related functions could also sharply raise gas fees, he added.

Public-key exposure methods and response strategies differ by blockchain. Bitcoin can hide a public key behind a hash until a transaction takes place, depending on the address type. Typical SegWit addresses provide an additional defensive layer by not directly exposing the public key until spending. Taproot addresses, by contrast, expose the public key directly on-chain. Not reusing addresses may help reduce exposure, but once a Bitcoin transaction occurs, the public key is disclosed with transaction data, and attackers can observe it while the transaction remains in the mempool. Using a new address alone therefore cannot eliminate all risks.

Some blockchains are considering countermeasures including adopting quantum-resistant cryptography. NEAR Protocol has enabled users to add a FIPS 204-based key, a quantum-resistant standard, to an existing mainnet account with a single transaction. Ethereum is also reviewing a post-quantum security roadmap that uses hash-based signatures and zero-knowledge proof aggregation. Ethereum co-founder Vitalik Buterin proposed in March 2024 a plan that assumes large-scale asset theft, in which a hard fork would roll back blocks and halt regular wallet transactions, then use zero-knowledge proofs for seed phrases to recover funds. Max Shannon of Bitwise Europe said the idea is not yet an execution plan with a specific timetable.

Solana is also devising a separate response plan. Solana Foundation Vice President of Technology Jacob Creech said that even if the signature system is neutralised, users could prove via hash-based proofs that they possess a secret seed and migrate their accounts to a safer signature system. A proof of concept has been carried out, but a network upgrade has not reached the stage of being deployed.

No cases have been confirmed so far in which AI actually broke existing cryptographic systems and stole cryptocurrency. If the threat becomes real, responses may be difficult to resolve by changing wallet addresses alone. That is because public-key exposure structures, wallet infrastructure, and network signatures and consensus mechanisms would all need broad inspection. The industry’s task is to avoid overreacting to an unverified threat while preparing for a scenario in which the foundations of cryptography are shaken. As AI’s mathematical capabilities advance rapidly, the security debate is expanding from simple wallet management to longer-term responses across blockchain infrastructure.