According to Woofun AI, Anthropic released research on AI-optimized quantum circuits, a development that directly triggered a fresh reassessment of the security of Bitcoin and traditional financial infrastructure.
The study used the Claude model to complete a reconstruction of the elliptic curve point-addition quantum circuit within 53 hours, significantly lowering computational costs.
This technical breakthrough not only touches the core defense mechanism of Bitcoin signatures, but also forces the banking industry to accelerate scrutiny of the security of its cryptographic connections, marking the shift of post-quantum migration from theoretical discussion into an urgent phase of engineering practice.
On the technical details, Claude Mythos 5 used deep mathematical derivation to fundamentally optimize the divsteps algorithm, sharply reducing Toffoli gate consumption.
In point-addition operations in quantum computing, the most expensive step is modular inversion.
Traditional computers use the divsteps algorithm with very high efficiency, but in quantum circuits, because computation must be reversible, each branch choice must be stored as a record for later cleanup, consuming large numbers of qubits.
Claude found that these records do not need to be stored explicitly, because after each step the remaining number pair itself contains enough information to reconstruct the previous branch path.
Although the initial rules tried on 256-bit numbers all failed, a breakthrough was eventually achieved through derivation of algebraic identities and the use of Lean to formally prove the core properties of divsteps.
The design consumes about 12% more qubits than the previous most qubit-efficient scheme, but the number of Toffoli gates fell to about one-seventeenth.
The same automated iterative method was applied to another 7 tasks, one of which produced the largest improvement by reducing the product of qubits and gate count by a factor of 39.
By comparison, an estimate in quantum chemistry was reduced from 50 trillion Toffoli gates to 340 million, a feat that took multiple research teams 8 years of relay work, while this optimization required only 53 hours of continuous operation.
However, this technical breakthrough has clear limitations and must be assessed together with a complete attack scheme and different hardware constraints.
First, the optimization targets only a single module; the full process from inputting a Bitcoin public key to recovering the private key also involves other computational steps, error-correction overhead and hardware architecture, and only by putting the new circuit back into the complete scheme for recalculation can the final proportion of resource savings be determined.
Second, the one-seventeenth comparison baseline is the scheme by Luo et al., which requires about 69.5 million Toffoli gates and 838 logical qubits, while Claude's scheme requires about 4 million gates and 935 logical qubits.
If the opponent is changed, the conclusion changes: Schrottenloher's scheme this year requires only about 1.9 million gates but needs 1,446 logical qubits.
By contrast, Claude's circuit has 2.1 times the gate count of Schrottenloher's scheme but about 35% fewer qubits.
Whether gate count or qubit count is more critical depends on specific hardware constraints, which shows that the new scheme is more of an optional design point than an all-round victory.
In addition, the circuit allows a failure rate of up to 2×10⁻⁴ per point addition in exchange for resource savings, and the Lean proof covers only part of the properties of divsteps.
The verification process relies on simulation checks by an automated evaluation program and was not run on real quantum hardware, and Anthropic also noted that Claude was initially conservative and needed researchers to guide it to open up its thinking.
Data compiled by Woofun AI shows that estimates of the resources needed to break secp256k1 have declined exponentially over the past few years.
In 2017, Microsoft researchers estimated that about 2,330 logical qubits and 2.3×10¹¹ Toffoli gates would be needed.
At the end of March this year, Google's quantum AI team, with collaborators including the Ethereum Foundation's Justin Drake and Stanford University's Dan Boneh, proposed two circuits: one with fewer than 1,200 logical qubits and 90 million gates, and another with fewer than 1,450 logical qubits and 70 million gates, both requiring fewer than 500,000 physical qubits and running in about 9 minutes in specific scenarios.
In September, IonQ further estimated that a break could be completed in about 26 days using about 20,000 trapped-ion qubits.
From 2017 to March this year, the interval between estimate reductions was nearly 9 years; by September, IonQ's new estimate came only half a year later, showing that the pace of technological progress is accelerating.
Although estimates keep falling, current quantum hardware capability remains limited, but research at the intersection of AI and quantum cryptography has become a new focus.
At present, quantum computers have only a few hundred to a few thousand qubits, and demonstrated logical qubits range from a few dozen to nearly a hundred.
According to Project Eleven's Q-Day award records in April this year, the largest elliptic curve key broken in a public demonstration on real quantum hardware was only 15 bits.
Another paper this year, ECDSA.Fail, emphasized the importance of open automated research.
Project Eleven pointed out that the next round of challenges will focus on the intersection of frontier AI models and quantum cryptanalysis.
Anthropic's research and the estimates by Google and IonQ cannot be directly added together: the former is a single-module design, while the latter two are estimates of resources for an entire attack, but together they point to a rapid decline in attack costs.
Bitcoin faces a unique migration dilemma because it lacks a centralized decision-making mechanism, and BIP proposals progress slowly.
BIP-360 was merged into the official proposal repository on February 11 and remains a draft, offering only a new address type for users to migrate voluntarily.
The more radical BIP-361, proposed by developer Jameson Lopp and others, envisions eventually ceasing to recognize old signatures, which has sparked a property-rights debate over whether long-dormant old coins should be frozen.
BIP-360 co-author Ethan Heilman estimates that even if started now, a complete Bitcoin migration to a post-quantum system would take about 7 years.
The risk exposure is huge: about 6.04 million bitcoins have public keys already exposed, accounting for 30.2% of issued supply.
By contrast, Ethereum has moved more proactively: it set up a dedicated post-quantum security team in January this year, and Vitalik Buterin published a roadmap in February, showing stronger organizational coordination.
Banks' response strategies are relatively clear but complex to execute, and market views are divided.
NIST's draft transition timeline stipulates that algorithms such as RSA-2048 with 112-bit strength will be deprecated after 2030, and all public-key algorithms vulnerable to quantum attack will be banned after 2035.
SWIFT plans to launch SwiftNet 8.0 with post-quantum cryptography support in 2027.
However, a BankInfoSecurity commentary pointed out that most banks cannot clearly explain their cryptographic deployments.
BlackRock disclosed quantum risk for the first time in its May 2025 Bitcoin ETF registration filing, when the fund had about $64 billion in net assets.
Bloomberg Intelligence analyst James Seyffart reminded that such disclosures are intended to list all risks.
On price, Glassnode analyst James Check believes the price weakness is mainly due to selling by long-term holders, while Bitcoin developer Matt Corallo pointed out that if quantum risk were the main cause, Ethereum should outperform Bitcoin.
Banks can advance migration through unified management decisions on budgets and scheduling, whereas Bitcoin's fees are dispersed and no one can guarantee synchronized upgrades, leaving the fate of some old assets uncertain.
This is a key moment, following AI-accelerated quantum algorithm optimization, when post-quantum migration shifts from technical feasibility to a social engineering challenge.
Comments