Claude Spent a Week on Cryptanalysis; Humans Needed Nearly a Month to Verify It
On July 28, Anthropic published a set of hard results: Claude Mythos Preview helped researchers improve a key-recovery attack against the HAWK post-quantum signature scheme and found a new attack on 7-round AES-128.
Let's set the boundaries first. It did not break full AES as used in production systems today, nor did it make internet encryption obsolete overnight. What's truly worth watching is that, for the first time, the speed of research and the speed of verification have diverged this dramatically.
HAWK: From 2^64 Down to 2^38
HAWK is a post-quantum digital signature candidate. Claude found a non-trivial automorphism in the lattice structure it uses, which accelerates an enumeration-based key-recovery attack.
For HAWK-256, the expected attack cost dropped from about 2^64 to 2^38. That is no small tweak. The effective security strength was nearly cut in half.
That said, the attack is still exponential-time, does not affect other NIST post-quantum signature candidates, and certainly doesn't overturn lattice cryptography as a whole. HAWK hasn't been deployed in production; its research purpose is precisely to surface weaknesses before standardization.
The research process is interesting: multiple worker agents collaborated in sandboxes with access to Python, Sage, and cryptographic literature. One worker judged the key idea infeasible and gave up too early; another kept pursuing it and found the full exploitation path. Only after exchanging notes did the two converge. About 60 hours total.
This is not "copy one model ten times and truth emerges." The real productivity of multi-agent work comes from mechanisms for disagreement, record-keeping, review, and continued exploration.
AES: Mobius Bridge Compresses One Step by 256x
The second result targets 7-round AES-128. Full AES-128 has 10 rounds, so this is a reduced-round study subject in cryptanalysis.
Claude proposed a fingerprinting method called Mobius Bridge, replacing a stage that originally required enumerating 256 values with a guess-invariant fingerprint. That step's workload drops 256x, and combined with other optimizations, the overall attack is roughly 200 to 800 times faster than the previous best.
The numbers are impressive, but the practical threat is minimal: the attack requires about 2^105 chosen plaintexts, making it entirely impractical, and it does not apply to full 10-round AES.
The model initially refused the task, arguing that AES analysis is mature and progress unlikely. Researchers repeatedly corrected it, telling it not to chase low-hanging fruit. Within three days the model produced hundreds of millions of tokens; results formed in about a week, totaling nearly a billion tokens. The API cost of the single discovery was roughly $100,000.
The Real Bottleneck Is Now "Human Verification"
Two Anthropic researchers spent hundreds of hours brushing up on cryptography to verify the AES result. Claude produced the attack in a week; humans needed nearly a month to trust it.
That sentence carries more weight than any benchmark score.
AI for Science is often framed as "models helping scientists," with the assumption that scientists remain in control of judgment. But when models push past the boundaries of human knowledge, researchers must simultaneously verify correctness, novelty, and real-world impact. As generation speed keeps climbing, the verification queue will overflow first.
Anthropic has partnered with academia on CryptanalysisBench and followed responsible disclosure, coordinating in advance with the HAWK authors, relevant NIST communities, and government and industry partners. The direction is right, but far from sufficient. If a future model finds a practically exploitable flaw in a production cryptosystem, the cadence for disclosure, reproduction, and patching must be agreed on in advance, not improvised on the spot.
My conclusion is simple: this time nothing "broke the internet," but it proves frontier models can already find new things in mathematical objects that experts have scrutinized for years. The research harness has matured. What will bottleneck the field next is the verification harness.
Original sources
- Anthropic research post: https://www.anthropic.com/research/discovering-cryptographic-weaknesses
- Technical demo repository: https://github.com/anthropics/cryptography-research-demo
- Related coverage: https://the-decoder.com/anthropic-says-its-mythos-model-found-vulnerabilities-in-cryptographic-algorithms-that-secure-the-internet