AI Discovery Forces Immediate Withdrawal of HAWK PQC Signature Standard

The digital signature scheme HAWK , which had been under consideration as a candidate for official post-quantum cryptographic (PQC) standardization, has been withdrawn after a newly identified security weakness reduced its effective protection.

Security experts said the issue amounts to a vulnerability that effectively halved the algorithm’s cryptographic strength. The weakness was linked to previously unknown structural properties inside the problem HAWK relies on, described as a Lattice Isomorphism Problem.

AI-assembled tools exposed new symmetries According to the findings described by security experts According to the findings described by security experts, the flaw was uncovered using an artificial intelligence model that combined existing mathematical techniques rather than inventing an entirely new method. The model synthesized known tools to detect previously unknown symmetries in the underlying problem, enabling an attack path that had not been documented before. Experts said the AI system carried out much of the work with minimal human cryptographic expertise. They described it as automating a process that included a literature review and a verification pipeline, and then using that workflow to execute the attack. Mitigation existed, but the cost made HAWK uncompetitive The vulnerability could, in theory, be addressed by doubling key sizes, according to the assessment. However, the same assessment said that doing so would create enough computational overhead to make HAWK unattractive compared with existing PQC alternatives. That trade-off—raising parameters to restore security at the That trade-off—raising parameters to restore security at the expense of speed and efficiency—was central to the decision to withdraw the scheme from the standardization track. The result is that HAWK will no longer be evaluated alongside other post-quantum options in that selection pipeline. Why the episode matters for PQC standardization Experts emphasized that the current findings do not translate into an immediate break of production-grade systems. The more direct consequence is for how candidates are assessed: a vulnerability that cuts security margins and requires large parameter increases can disqualify an algorithm even if a theoretical fix exists.

The episode also points to a broader risk highlighted by the researchers: AI agents can accelerate cryptographic weakness discovery by systematically connecting known but separate analytical methods. In practical terms, automating the search, validation, and execution steps may reduce the time and cost required to find attacks against complex protocols.

Security experts said the incident signals a change in the threat landscape surrounding cryptographic standards, as defenders may need to assume that more of the early-stage discovery work can be automated and scaled.