Claude AI cracks post-quantum security test in under four hours

Claude AI has cracked a post-quantum encryption test scheme and discovered a 200 - 800 times faster attack on a reduced-round version of AES-128.

2

The breakthrough, achieved using Anthropic's Mythos Preview AI, comes just as the field of post-quantum cryptography is gaining critical attention ahead of NIST standardization. So, for those who don't know, AES-128 is the encryption standard used throughout modern computing, and full AES-128 performs 10 rounds of mathematical transformations. Claude attacked a research version of HAWK-256, a digital signature system designed to remain secure even against quantum computers. The system is currently being tested by the National Institute of Standards and Technology (NIST).

Claude attacked the research version of HAWK with only seven rounds and discovered hidden symmetry in the HAWK-256 lattice-based algorithm, reducing its effective key strength and enabling a full key recovery in about 3 hours and 42 minutes on a 96-core server. Separately, Mythos found a way to streamline a seven-round AES-128 attack by eliminating a 256-way guessing step, though both remain infeasible for real-world systems.

Anthropic says neither result threatens production software, as HAWK-256 and the results gathered from the testing are used to gauge cryptographic resilience, which has proved beneficial as Claude has demonstrated how AI models can be used to evaluate encryption schemes before they are deployed, or to simply discover weaknesses that humans have overlooked.

Frequently Asked Questions

TweakBot answers common questions about this news using TweakTown's own coverage from this page and related content from our archive. Tap a question to reveal the answer, or type your own below.

Question #1

Which specific version of HAWK-256 did Claude attack, and how many rounds did it have?

Claude attacked a research version of HAWK-256 that had only seven rounds. The attack recovered the full key in about 3 hours and 42 minutes on a 96-core server.

Answered

Question #2

How long did Claude take to recover the full key for the weakened HAWK-256 instance and what hardware was used?

Claude recovered the full key in about 3 hours and 42 minutes. The attack ran on a 96-core server.

Answered

Question #3

What hidden property did Claude discover in the HAWK-256 lattice algorithm that reduced its effective key strength?

Claude discovered a hidden symmetry in the HAWK-256 lattice-based algorithm that reduced its effective key strength. This symmetry enabled a full key recovery in about 3 hours and 42 minutes on a 96-core server.

Answered

Question #4

Does Anthropic say the Claude results pose a threat to production HAWK-256 deployments or current post-quantum standards?

Anthropic says neither result threatens production software, noting the HAWK-256 test and the findings are used to gauge cryptographic resilience. The company framed Claude's work as beneficial for evaluating encryption schemes before deployment and for discovering weaknesses humans may have missed.

Answered

Have a question not listed here? Ask below and TweakBot will answer it.

Your questionAsk

Answering your question×

TweakBot is analyzing your question…

Could not get an answer

TweakBot is generating an answer

Read More

Was TweakBot helpful?Thanks for the feedback!

The company said the AI conducted the bulk of the research autonomously, with human input limited to oversight and verification. The cost? Roughly $100,000 in API usage for both projects, for a total of $200,000 in API costs.