On October 16, Anthropic published two pieces of research exploring how its Claude model could help investigate cryptographic problems. One focused on designing more efficient quantum circuits, including for mathematical operations relevant to the quantum threat to elliptic-curve cryptography. The other investigated the origins of parameters used in standardized elliptic curves, a question that has attracted scrutiny since the National Security Agency (NSA) supplied the original values without a public explanation.
Together, the studies show how AI can assist with quantum-circuit optimization and cryptographic research. They do not establish that Bitcoin is closer to being broken by a quantum computer, nor do they prove that the historical NIST curve parameters were free of every possible form of manipulation.
Part One: Claude Designs Leaner Quantum Circuits
Quantum circuits consist of sequences of operations called gates. Researchers seek to reduce the number of gates and qubits required to perform a calculation because these resources affect the cost and feasibility of running quantum algorithms.
In its October 16 research post on quantum-circuit design, Anthropic said it gave Claude Mythos 5 plain-language specifications for eight computational tasks spanning chemistry, optimization, quantum simulation, number theory and numerical computing. An automated evaluator checked each proposed circuit for correctness and scored its resource requirements.
Claude then worked in a feedback loop, proposing a design, receiving the evaluator’s results and revising its approach. Anthropic said the process ran without human input between rounds, although researchers supplied the initial instructions and occasionally prompted the model to explore different approaches.
The company reported improvements across all eight tasks, with some results reaching an order of magnitude or more. The work was led by Andi Gu through Anthropic’s STEM Fellows Program, with mentorship from Alek Kemeny and collaboration with Harvard researchers Susanne Yelin and Mikhail Lukin, who is also affiliated with QuEra Computing. These are results reported by Anthropic; independent replication would provide a separate basis for assessing their robustness.
The 17x Result Relevant to Bitcoin
One task involved elliptic-curve point addition, a mathematical operation used in quantum algorithms for solving the elliptic-curve discrete logarithm problem. According to Anthropic, Claude worked for 53 hours on a reversible implementation of an algorithm called divsteps for modular inversion. The resulting design avoided storing the full history of conditional operations, reducing the number of gates needed to execute the circuit.
One design used 12% more qubits than the smallest previous design but required 17 times fewer gates. That trade-off matters: it represents a substantial reduction in gate count, not a reduction in every resource measure.
The connection to Bitcoin is mathematical. Bitcoin uses the secp256k1 elliptic curve for its ECDSA and Schnorr signature systems. Their security relies on the difficulty of recovering a private key from its public key. A sufficiently capable, fault-tolerant quantum computer running Shor’s algorithm could threaten that assumption by solving the elliptic-curve discrete logarithm problem.
The Bitcoin developer documentation describes Bitcoin’s use of secp256k1, while Anthropic’s quantum-circuit research explains why point addition is a central operation in quantum algorithms for elliptic-curve discrete logarithms.
A more efficient point-addition circuit could contribute to lower resource estimates for a future quantum attack. But the 17x figure applies to one circuit design, not to a complete attack on Bitcoin. Anthropic did not calculate that its result reduces the total cost or runtime of breaking Bitcoin by 17 times.
That distinction is essential. A full quantum attack would depend on the circuit design, the number and quality of physical qubits, error-correction overhead and the resources needed to execute the complete algorithm. A gain in one component does not, by itself, establish when a cryptographically relevant quantum computer might become practical.
Anthropic’s post presents point addition as a standard benchmark for quantum algorithms. The company does not claim to have broken encryption, derived a Bitcoin private key or demonstrated an attack on secp256k1. The Bitcoin connection is relevant technical context, not a claim made by the company.
For more background, read The Crypto Times’ report on Google’s quantum research and what it means for Bitcoin.
Part Two: Claude Traces the Origins of NIST Curve Seeds
Anthropic’s second investigation addressed a different cryptographic question: how the seed values for several standardized elliptic curves were originally generated.
NIST standardized elliptic curves in its digital-signature standards, including P-192, P-224, P-256, P-384, and P-521. Their published parameters include seed values used in defining the curves. The NSA supplied those values without a public explanation of how they were chosen, prompting questions about whether the process could have allowed deliberate weaknesses to be introduced.
Those concerns gained attention after the discovery of a backdoor in Dual_EC_DRBG, a separate random-number-generation standard. However, suspicion about the NIST curve seeds was not evidence that those curves contained a backdoor.
In its October 16 report on ECDSA curve origins, Anthropic said Claude investigated the published seed values without being given the suspected explanation. The model hypothesized that the seeds might have been generated by an older pseudorandom number generator whose output patterns could be recognized.
Claude built a scanner to look for those patterns and identified characteristics consistent with lrand48, a pseudorandom number generator found in older C libraries. Anthropic said the model recovered the generator’s internal state, using brute force for some binary-curve seeds and the Lenstra–Lenstra–Lovász (LLL) lattice-reduction algorithm for truncated outputs associated with prime-field curves.
The reconstructed states pointed to Unix timestamps from 1997 and 1998 for several prime-curve seeds. Anthropic said Claude reproduced the seed values in under half an hour and published code intended to let others reproduce the results.
The finding provides a concrete, testable explanation for the origins of several seed values. It does not mean Claude discovered a backdoor; rather, it offers evidence against the theory that those seeds were deliberately engineered through a secret cryptographic process.
Anthropic concluded that the findings make a deliberate backdoor “much less plausible” and said the parameters could be regarded as secure. The company also stated that the work did not identify a cryptographic break, a new attack or a weakness in the ECDSA curves.
What Remains Unresolved
The result should not be treated as a complete explanation of every parameter in every standardized curve. Anthropic acknowledged that some details remain unexplained. Its researchers could not recover one of the binary-curve seeds, and the origins of several curve base points remain unknown. The researchers also observed that some candidate seeds appeared to have been skipped during the original generation process, without establishing why.
Those gaps do not establish that a backdoor exists. They do mean the finding should be described as a reproducible explanation for several seed values, rather than definitive proof that every aspect of the original curve-generation process has been resolved.
Anthropic said it would donate the $36,864 bounty associated with the discovery to the Internet Security Research Group. The company’s report also acknowledged Jerry Solinas, the NSA mathematician associated with the curve design, who passed away in 2023.
For broader technical context, NIST’s Digital Signature Standard describes the standards governing digital signatures, while its guidance on elliptic-curve domain parameters covers recommended curves.
Why the NIST Findings Do Not Directly Apply to Bitcoin
The crucial distinction for crypto readers is that Anthropic investigated NIST and related standardized curve parameters, not Bitcoin’s secp256k1 curve.
Bitcoin’s ECDSA signatures use secp256k1, a different curve with different parameters. The Bitcoin developer documentation identifies secp256k1 as the curve used for Bitcoin public keys. Consequently, the findings about how the NIST seed values were generated should not be presented as a direct security assessment of Bitcoin’s curve.
The NIST research may still matter to the wider digital-asset ecosystem. P-256 and other standardized elliptic curves are used in many cryptographic applications, including digital-signature systems and internet security infrastructure. A clearer account of their parameter origins is relevant wherever those curves are deployed.
The result is therefore broader than Bitcoin, but less direct than a claim that Claude has resolved a security problem in Bitcoin itself. It helps explain a longstanding question about certain widely used cryptographic standards while leaving Bitcoin’s secp256k1 security outside the scope of that investigation.
The two studies also address different types of problems. The quantum-circuit work explores how to reduce the computational resources required by certain quantum algorithms. The curve-origin research uses pattern recognition, algorithmic reconstruction and mathematical techniques to investigate how historical parameters were generated.
Neither study establishes that Claude has found a new way to recover Bitcoin private keys using classical computing, or that the curve-origin results make Bitcoin more or less vulnerable to a future quantum attack.
What the Two Studies Mean for Crypto
For Bitcoin holders, neither result demonstrates an immediate change in security. Anthropic’s quantum-circuit work is a circuit-design advance, not a practical attack on secp256k1. Its curve-origin research examines different curves and explicitly does not claim to have found a cryptographic weakness.
The broader lesson is that AI is becoming a tool for research into cryptography and quantum computing. It can help researchers explore alternative circuit designs, test mathematical hypotheses and reconstruct the origins of cryptographic parameters. That capability could influence future security assessments, but each result still needs to be evaluated within its technical scope and independently checked where possible.
The quantum research also reinforces the rationale for continued work on post-quantum preparedness. Bitcoin developers have proposed changes intended to address specific quantum-related risks. For example, BIP-360 proposes a Pay-to-Merkle-Root output type that removes Taproot’s quantum-vulnerable key-path spending mechanism. It remains a proposal, not a completed network-wide migration to post-quantum signatures.
The Crypto Times has also covered Coinbase’s post-quantum security priorities for Bitcoin, including the technical and coordination challenges involved in preparing digital-asset infrastructure for future cryptographic threats.
Anthropic’s findings do not establish a deadline for that transition. They add to the research that developers and security teams can consider while assessing future risks. The Crypto Times does not forecast when, or whether, a quantum computer will become capable of breaking Bitcoin’s cryptography.
The central distinction remains: Claude has produced a more gate-efficient design for one quantum-circuit benchmark and an explanation for the origins of several NIST curve seed values. Neither result demonstrates that Bitcoin’s cryptography has been broken, and the NIST findings should not be mistaken for a direct assessment of secp256k1.
Also Read: Fed Chair Warsh Gives No October Rate Signal, Warns AI Boom May Fuel Inflation
