Anthropic's AI Discovers Attacks on Post-Quantum Cryptography Bitcoin May Adopt

Editorial illustration for: Anthropic's AI Discovers Flaws in Post-Quantum Cryptography, Raising Bitcoin Security Questions

In brief

  • Anthropic's Claude Mythos Preview discovered novel attacks on weakened AES and HAWK post-quantum schemes.
  • AI-generated attack accelerated AES decryption 200–1,000 times faster than known methods.
  • Bitcoin currently uses ECDSA but developers are exploring lattice-based post-quantum alternatives.
  • Researchers confirmed modern encryption remains secure; attacks targeted deliberately weakened research variants.
  • Bitcoin may need to evaluate post-quantum security against both quantum and AI threats.

The Discovery

Anthropic's Claude Mythos Preview model reportedly uncovered previously unknown attacks against deliberately weakened versions of the Advanced Encryption Standard used in research environments. The attacks targeted deliberately weakened cryptographic systems rather than production-grade implementations, meaning real-world AES deployments face no immediate threat. Anthropic stated that modern AES remains secure.

The significance lies not in breaking current encryption, but in demonstrating that frontier AI models are becoming capable of producing original cryptographic research. Researchers say this accelerates the timeline for security concerns that were previously theoretical.

Bitcoin's Cryptographic Roadmap

Bitcoin currently relies on ECDSA to authorize transactions. Cryptographers have warned that sufficiently powerful quantum computers could eventually break ECDSA, prompting developers to explore alternatives. Bitcoin developers have begun working on migrating to post-quantum signature schemes, with lattice-based cryptography emerging as a leading candidate.

Anthropic's findings complicate this migration. If AI systems continue advancing—and if they keep discovering novel weaknesses in cryptographic primitives—Bitcoin's shift to post-quantum schemes may solve one problem only to create another.

The Dual-Threat Timeline

Neither Bitcoin nor today's internet encryption is in immediate danger from the discovered attacks. Researchers stressed this point repeatedly. But the implication is sobering.

Peter Van Valkenburgh, Executive Director of Coin Center, framed the scenario this way: "Coin Center Executive Director Peter Van Valkenburgh joked about the "funniest timeline," in which Bitcoin upgrades to a lattice-based post-quantum signature scheme only for advanced large language models to break the underlying mathematics before quantum computers ever manage to pose a threat."

The researchers' own assessment is more measured. If AI systems continue accelerating mathematical discovery, developers of Bitcoin and other blockchain networks may eventually need to evaluate post-quantum security against both quantum and AI cryptanalysis. It's a problem that didn't exist five years ago. Now it demands attention from protocol engineers building systems meant to last decades.