Could AI or quantum computers break Bitcoin’s encryption?

In late July 2026, artificial intelligence startled cryptographers. A model found an unseen mathematical symmetry inside HAWK, an experimental signature scheme up for standardization, cutting its security in half. That prompted Google cryptography specialist Sophie Schmieg to deliver a blunt verdict: “Basically with this paper, HAWK is dead.”

When software can dissect complex math like that, it is natural to worry about digital vaults. A listener asked why cryptocurrency wallets stay secure as AI and supercomputers advance, whether a supercomputer running AI could break into wallets with ease, and what protects digital assets if those tools end up in the wrong hands. To find out, we trace the route from the immense math guarding every coin, through the operational blunders where money actually disappears, to the quantum horizon.

The system behind Bitcoin came together in 2008, adapting concepts for signed ledger chains that date back to 1991. It relies on asymmetric public-key cryptography—a setup that lets you authorize a spend with a secret key without ever exposing it. It uses elliptic-curve signatures called ECDSA with secp256k1 parameters, alongside a digital fingerprint known as double SHA-256.

That design relies on a 256-bit key space, which is simply the complete inventory of keys a thief would have to check. Trying to guess one means sorting through two to the power of 256 options—a number 78 digits long that classical machines have no hope of brute-forcing. Yet if blind guessing is out of the question, what keeps an AI from finding a clever shortcut around the equations?

Machine learning thrives on finding patterns, but strong encryption exists to eliminate them. It produces entropy—pure mathematical noise with zero structure for a model to latch onto. A December 2025 study led by Bandyopadhyay concluded that using deep learning to recover keys from complete modern ciphers is practically infeasible, since the equations leave no predictable signals.

In July 2026, alarming claims that AI had cracked the Advanced Encryption Standard also fell flat. Anthropic clarified that its attack only worked on a stripped-down 7-round test version of AES-128, the global standard for scrambling data. That left the full 10-round cipher safe. Even that limited victory took heavy human labor. Researcher Nobuyuki Sugio observed that AI-generated cryptanalytic code often carries logical errors needing manual fixes. As cybersecurity author Marin Ivezic pointed out, the AI ran for tens of hours, but two human specialists spent hundreds of hours checking the work.

Raw computing power hits that exact same obstacle. A standard computer can run millions of SHA-256 checks each second, but cryptographic hashing is strictly a one-way process. Generating a digital fingerprint takes a split second, yet reversing it would take longer than the age of the universe. Even a supercomputer churning through billions of guesses every second is operating in the dark, and all those calculations are useless unless someone can actually spot the exposed front door.

This protection extends far beyond digital coins. Bandyopadhyay pointed out that cryptography serves as a bedrock of modern economic life, and Bitcoin shares these very tools with about 8,500 federally insured financial institutions across the United States. If artificial intelligence managed to crack these equations, international banking would collapse that same afternoon. So if digital assets aren't being lost to broken math, where is the real vulnerability coming from?

Where systems actually fail isn't in broken encryption—it's sloppy key handling and everyday coding mistakes. During its July 2026 evaluations, Anthropic found Claude got inside three outside corporate setups. It didn't untangle any algorithms or derive private keys. The model simply walked through open doors, taking advantage of unguarded endpoints and weak passwords left on the public web.

Real asset thefts follow that exact playbook. On September 6, 2026, an attacker targeted a software flaw in Liquid, a Bitcoin sidechain network designed for fast transfers. Programmers had linked hash fields back-to-back without boundary markers, leaving a subtle byte-alignment defect that automated scans had missed. That mistake let the intruder generate 4,000 unbacked coins and pull out about 4,000 bitcoin through SideSwap, an automated bridge participant that had left its approval key hooked directly to the internet.

No mathematical locks gave way. In a bizarre twist, the intruder wrote directly onto the blockchain: "we are whitehats. contact us on chain." After talks, they handed back 3,400 bitcoin, leaving roughly 602 unaccounted for. The problem was human implementation, not shattered math.

Because software is rarely perfect, real asset protection uses layered defense. Wallets store the keys that control funds, but institutional custodians isolate them with air-gapped cold storage, keeping hardware entirely disconnected from the web. Networks also rely on formal community standards known as Bitcoin Improvement Proposals, like BIP 11 for multi-signature transactions, where spending requires multiple separate keys like co-signers on a vault. Wire-level encryption shields node broadcasts, while offline custody acts as a buffer that buys time to spot unauthorized movement. Still, even if standard computers cannot break those shields, a deeper question remains: could a totally different computing architecture someday threaten the underlying math?

If classical supercomputers cannot break modern encryption, quantum computing poses a genuine theoretical threat. Technical expert Sultan Meghji highlighted that conventional computers cannot solve the discrete-logarithm math behind encryption in useful time. Yet back in 1994, years before Bitcoin existed, mathematician Peter Shor introduced Shor’s algorithm—a quantum blueprint capable of solving that math directly, cracking keys quickly and efficiently rather than taking billions of years.

That does not mean quantum mechanics tears down every defense. Even if Shor’s algorithm threatens elliptic curves, Bitcoin’s SHA-256 hashing holds firm. Running Grover’s algorithm—a quantum search technique used to speed up brute-force guessing—would trim SHA-256's workload from 2^256 down to 2^128 operations. While that is weaker than the classical standard, 2^128 operations remains a formidable tally, keeping the hash well out of reach.

Supporters also highlight an inherent defense: address hashing. Rather than displaying a raw key, the protocol creates an alphanumeric address by hashing it, keeping the master key inside a sealed, opaque envelope until you pay. Routing coins to that hash hides the public key while the balance sits untouched. With the key locked behind that one-way fingerprint, there is no exposed front door. A quantum machine has nothing to examine until the owner publishes the full key to authorize a spend.

For researchers who view practical quantum threats as distant, today's physical hardware remains far short. State-of-the-art processors like Google’s Willow chip feature only around 105 qubits. A company spokesperson confirmed that “the Willow chip is incapable of breaking modern cryptography.” Engineering quantum systems capable of challenging 256-bit curves remains years away. That shifts the real test to whether the network can agree to replace its locks before a relevant quantum machine arrives.

Whether the network can patch itself in time pits quantum engineering against human consensus. Skeptics point out that the danger is already written across the ledger. According to the draft proposal BIP 361, over 34 percent of all circulating coins have already exposed their public keys on-chain. A Google-commissioned study estimated that puts around 6.9 million BTC inside address types that would be vulnerable if a cryptographically relevant quantum computer existed today.

That backlog makes critics nervous because Bitcoin moves slowly. Protocol upgrades historically take two to four or more years from proposal to activation. With road maps attributed to McKinsey warning that quantum machines could arrive as early as 2027 to 2030, the window feels tight. Yet when BIP 361 authors proposed locking down inactive coins, pushback was fierce, with community members branding the move “highly authoritarian and confiscatory.”

Cryptographers counter that this alarm overlooks the gap between experimental chips and full-scale attack computers. Cracking Bitcoin’s elliptic curves would require about 2,330 logical qubits. The distinction matters because fragile physical qubits need heavy error correction to form a single stable logical qubit. Accounting for that noise raises the bar to between 900,000 and several million physical qubits, which is why expert consensus places the earliest plausible threat between 2030 and 2035.

That timeline gives developers room to adapt. On August 13, 2024, NIST published official post-quantum standards like ML-DSA and SLH-DSA. Meanwhile, Bitcoin engineers drafted BIP 360, proposing an upgrade called Pay-to-Merkle-Root to eliminate long-exposure quantum risks. As a soft fork—a backward-compatible upgrade that reuses tested code from Taproot, an earlier Bitcoin protocol upgrade for complex spending scripts—it adapts proven Taproot code to strip away exposed key paths and shield balances. What remains to be seen is how digital networks historically manage when their mathematical locks finally reach an expiration date.

Networks survive new hardware through cryptographic migration—the habit of replacing aging ciphers before machines catch up, treating encryption as an evolving shield rather than a stone tablet. In January 1999, the Electronic Frontier Foundation and distributed.net proved that necessity by cracking a 56-bit DES key in about 22 hours. NIST responded in 2001 by selecting AES, and security moved forward.

Preparing for quantum computing follows that playbook. In 2025, NIST project leader Dustin Moody selected HQC as a mathematical backup, emphasizing the need for a fallback if primary defenses fail. On Bitcoin, BIP 360 was published on March 10, 2026, placing quantum protection on the road map. Protocol expert Wolfgang Amadeus Vitale expected social consensus around such a soft fork to take shape through 2026.

An AI on a supercomputer cannot crack crypto wallets because modern full-strength encryption leaves no patterns for machine learning to exploit. Federal standards researchers and security analysts emphasize that real theft stems from lost or exposed private keys, not broken math. That operational failure was how an intruder walked away with roughly 4,000 bitcoin before posting on-chain that "we are whitehats."

When Sophie Schmieg declared that “HAWK is dead,” it showed the process working. Researchers caught the flaw early, prompting the authors to withdraw the algorithm before it could ever be deployed. Bitcoin stays safe because its 256-bit mathematics gives classical machines no shortcuts, and because developers are already preparing new cryptographic locks well before the old ones give way.

Spend some time with Bitcoin to understand the scale of the math, and follow the National Institute of Standards and Technology; we have more stories on them, so you stay ahead as cryptography standards advance. That's it for today. Thanks for listening, and we'll catch you next time on Angle.