Anthropic has announced that its experimental AI model, Claude Mythos Preview, discovered a previously unknown cryptographic weakness in HAWK-256, a post-quantum digital signature scheme, in roughly 60 hours. The finding has renewed discussion about the future of cryptographic security and whether Bitcoin and other blockchain networks should accelerate preparations for an era of increasingly capable AI systems.
The company said the model identified an attack that effectively cuts HAWK-256’s security margin in half. Researchers also reported a significant improvement in an attack against a reduced-round version of AES-128. While both discoveries represent meaningful advances in cryptanalysis, experts emphasized that neither result compromises encryption used in today’s production systems.
What the research found
According to Anthropic, the AI worked with only occasional human guidance and completed the HAWK-256 discovery after approximately 60 hours of analysis. The company estimated the research consumed about $100,000 worth of API compute re
The key findings include:
- A new key recovery attack against HAWK-256 that substantially lowers its effective security level.
- A 200 to 800 times faster attack on a seven-round version of AES-128, not the full 10-round AES-128 standard.
- Advance disclosure of the findings to the algorithm’s designers and relevant government and industry organizations before publication.
Researchers stressed that HAWK remains a proposed post-quantum signature scheme rather than a widely deployed standard. Likewise, the AES result applies only to a deliberately weakened research variant rather than the encryption protecting banking systems, messaging platforms, and internet traffic today.
Bitcoin implications
The research nevertheless sparked fresh discussion within the cryptocurrency community because Bitcoin’s long-term security depends on robust cryptography. Some observers argue that increasingly capable AI systems could shorten the timeline for discovering weaknesses in cryptographic algorithms, making proactive upgrades more important.
However, the new findings do not threaten Bitcoin’s current cryptography. Bitcoin relies primarily on SHA-256 hashing and the secp256k1 elliptic curve signature system, neither of which was affected by Anthropic’s research. The HAWK attack targets an unrelated post-quantum signature proposal, while the AES result concerns a reduced-round academic variant rather than deployed encryption.
Even so, the work highlights how frontier AI models are becoming powerful research tools for both discovering vulnerabilities and strengthening future cryptographic standards. As AI capabilities continue to improve, cryptographers and the cryptocurrency industry may face increasing pressure to evaluate and adopt more resilient security mechanisms before theoretical risks become practical ones.
