Eight PQC signature candidates enter NIST's third round after tweaks deadline closes on 14 August 2026

NIST's additional post-quantum digital signature process entered its third round on 14 May 2026 with nine candidates, then lost one to AI-assisted cryptanalysis before the tweaks deadline closed on 14 August 2026^[1]. Eight algorithms now carry the field into a review phase expected to last approximately two years^[1].

How the field narrowed to eight

NIST announced the nine third-round candidates on 14 May 2026 after 18 months of second-round evaluation, releasing the selection rationale in NIST Internal Report 8610^[1]. Five algorithms - CROSS, LESS, Mirath, PERK, and RYDE - were cut at the end of round two for uncompetitive performance trade-offs or security vulnerabilities identified during that period.

The nine that advanced were:

• FAEST
• HAWK
• MAYO
• MQOM
• QR-UOV
• SDitH
• SNOVA
• SQIsign
• UOV

Submission teams had until 14 August 2026 to file specification updates and implementation fixes - what NIST calls "tweaks" - to address known cryptanalytic vulnerabilities^[1]. NIST has signalled that significant changes at this stage may indicate an algorithm is not mature enough for standardisation.

HAWK did not survive to the tweaks deadline. On 28 July 2026, Anthropic's Frontier Red Team published findings showing that Claude Mythos Preview had identified a structural flaw in HAWK's lattice in approximately 60 hours of work, at a compute cost of roughly $100,000^[1]. The attack halved HAWK's effective key strength, and the HAWK team subsequently withdrew the scheme from the process. The eight remaining candidates - FAEST, MAYO, MQOM, QR-UOV, SDitH, SNOVA, SQIsign, and UOV - are now the active field.

What the AI result means for the remaining eight

The Mythos findings covered two targets. The HAWK result demonstrated that a frontier model could locate a mathematical flaw that two years of expert human review had not surfaced. The second result concerned a seven-round reduced variant of AES: Mythos developed a technique it named the Möbius Bridge that made the strongest known theoretical attack on seven-round AES-128 between 200 and 800 times faster. That attack requires over 400 octillion chosen plaintexts and cannot reach the full ten-round cipher used in production systems, so no deployed encryption is affected.

Anthropic notified the HAWK designers in June 2026 and coordinated public disclosure with the NIST mailing list on 28 July. To help the broader research community measure AI performance in this domain, Anthropic partnered with researchers at ETH Zurich, Tel Aviv University, and the University of Haifa to release CryptanalysisBench, a benchmark for evaluating large language models on cryptanalysis tasks.

The process question

The HAWK episode has surfaced a structural question for NIST's review cadence. The agency's evaluation timeline is built around human cryptanalysts working over months; a frontier model probed HAWK's mathematical foundations in 60 hours for roughly the cost of a small research contract^[1]. Commentators on the NIST PQC forum, including Markku-Juhani Saarinen, have argued that AI-assisted cryptanalysis results should ship with machine-checkable proofs or working demonstrations against scaled-down targets to allow rapid independent verification.

NIST is planning to hold the 7th PQC Standardization Conference in late spring or early summer of 2027 in or near Gaithersburg, Maryland, where the eight remaining teams will present updates. Whether the agency adjusts its evaluation methodology to account for AI-speed cryptanalysis before that conference is the question the remaining candidates now face.