# Claude Mythos broke HAWK and the NIST post-quantum timeline may not survive it > Source: > Published: 2026-07-30 02:20:34+00:00 *Claude Mythos Preview did not break production encryption, and HAWK has not been shown publicly as withdrawn. The real story is still sharp: Anthropic says an AI-assisted run found a serious weakness in a NIST post-quantum candidate after two years of human review.* Two years of expert review missed it. An AI model found it in 60 hours. That is the story worth your attention. On July 28, 2026, Anthropic's Frontier Red Team published research showing that Claude Mythos Preview had found a stronger attack against HAWK, one of the nine candidates NIST advanced in May to the third round of its additional post-quantum digital signatures process. Anthropic said the attack cuts HAWK's effective key strength in half. That is not a small edit. According to Anthropic's paper, the expected cost of a full key recovery attack against the small HAWK-256 parameter set was thought to be 2^64, but Mythos demonstrated a path at 2^38. The attack still doesn't run in polynomial time, and Anthropic was clear that larger HAWK keys remain impractical to attack. No production software has to change today, because HAWK is a candidate scheme, not a deployed standard. Keep that distinction in view. The attack works by finding a previously unused symmetry, a nontrivial automorphism, in the lattice structure HAWK relies on. Anthropic said prior work had shown that efficiently finding such an automorphism would permit an attack, but had not shown whether one was actually reachable in HAWK's lattice. Mythos reached it. One Anthropic researcher worked with the model in a Claude Code-like harness, using Python and Sage alongside published cryptographic papers. The human operator had a theoretical computer science background, Anthropic said, but wasn't a lattice cryptography specialist. That part should make you pause. The model didn't replace cryptographers. It changed the search cost. Anthropic estimates the HAWK result cost roughly $100,000 in API usage, and the full discovery and verification process took about 60 hours. If you're designing infrastructure around long-lived signatures, that is the number that matters more than the headline. An attack surface that took years to probe by hand can now be explored by a research group with a serious cloud bill. ## HAWK is weaker, not publicly dead NIST's additional signature process was already under pressure before Anthropic published the result. On May 14, 2026, NIST announced in NIST IR 8610 that nine schemes had moved into the third round: FAEST, HAWK, MAYO, MQOM, QR-UOV, SDitH, SNOVA, SQIsign and UOV. The agency said this third phase is expected to last about two years, with another NIST PQC standardization conference planned for late spring or early summer 2027. The published record does not support saying HAWK has been withdrawn. Anthropic says it shared the attack with HAWK's authors in June and coordinated disclosure to the public NIST mailing list when the research was released. CyberScoop also reported that Anthropic notified the designers and briefed U.S. government and industry partners. That is disclosure. It is not withdrawal. Frankly, the story doesn't need that extra drama. HAWK was attractive because it offered fast signing and compact signatures compared with many post-quantum alternatives. Anthropic's finding says preserving the same security level would require doubling HAWK's key size, which would remove much of the appeal. Cloudflare's Bas Westerbaan and Christopher Patton had already argued on July 9 that organisations shouldn't wait for better post-quantum signature algorithms, because ML-DSA is the best all-around standardised option available now. HAWK's weakness makes that argument easier to believe. ## The AES result is different There is a second result in Anthropic's research, and it should not be mixed up with HAWK. Mythos also improved an attack on seven-round AES-128, a reduced research version of AES. Full AES-128 uses 10 rounds. Your bank and browser are not suddenly broken - nor are your messaging apps. This is where the original draft needed correction. The technique Anthropic says Mythos called the Möbius Bridge belongs to the reduced-round AES work, not to the HAWK attack. In that AES case, Mythos improved on a meet-in-the-middle attack by eliminating a 256-value lookup step, then added optimizations that made the attack 200 to 800 times faster depending on the runtime measure. Anthropic also said the attack remains completely impractical, requiring 2^105 chosen plaintexts under the prior work's assumptions. Still, don't wave it away. It took Mythos about a week to find the AES idea, then Anthropic researchers spent several hundred human hours validating it and preparing the paper. Later in the same post, Anthropic said two researchers needed nearly a month to gain confidence in the AES method. The bottleneck moves from discovery to verification. That is a real operational change for cryptography research. The migration clock is also real. A June 22, 2026 White House executive order directs federal agencies to transition high value assets and high impact systems to post-quantum key establishment by December 31, 2030, and to post-quantum digital signatures by December 31, 2031. NIST is also supposed to run a migration pilot by December 31, 2027. If you're responsible for systems that have to survive that transition, waiting for a perfect algorithm is not a plan. Use the standards that exist. Build for crypto agility. The lesson from Mythos and HAWK is not that NIST's process failed. It is that the review process now has a new participant, and it works faster than the old one. **Also read:** [Samsung's chip division posted a 250-fold profit surge as AI memory shortages rewrite the rules of the semiconductor market](https://startupfortune.com/samsungs-chip-division-posted-a-250-fold-profit-surge-as-ai-memory-shortages-rewrite-the-rules-of-the-semiconductor-market/) • [OpenAI confirms it is building a family of devices as Apple sues over alleged hardware secrets theft](https://startupfortune.com/openai-confirms-it-is-building-a-family-of-devices-as-apple-sues-over-alleged-hardware-secrets-theft/) • [Zcash sealed a $1.7 billion shielded pool after a four-year-old bug that could have printed unlimited coins](https://startupfortune.com/zcash-sealed-a-17-billion-shielded-pool-after-a-four-year-old-bug-that-could-have-printed-unlimited-coins/)