The point: AI systems improving at cryptanalysis could validate the robustness of new post-quantum algorithms rather than jeopardize them — provided the hard problems themselves are secure.
Cryptography is currently undergoing a historic transition from RSA and elliptic curve cryptography to post-quantum algorithms. Security researcher Matthew Green argues that AI-powered cryptanalysis could actually be beneficial in this phase.
Cryptographer Matthew Green from Anthropic’s AI security team describes the present moment as optimal for developing AI-powered cryptanalysis capabilities. The industry is in a historic transition phase: established public-key algorithms such as RSA and elliptic curve-based procedures (EC cryptography) are being replaced by new post-quantum algorithms to counter threats from future quantum computers.
Green argues that AI does not necessarily represent a risk in this context. If AI systems actually gain in cryptanalysis capabilities, this could even be advantageous under two conditions: first, if the new mathematical hard problems on which post-quantum algorithms like HAWK are based are actually secure. Second, more intensive AI-powered testing would lead to these security assumptions being validated more thoroughly.
The catch: this thesis assumes we do not live in Impagliazzo’s Minicrypt scenario — a hypothetical world in which certain cryptographic hard problems are fundamentally weaker than assumed. If, on the other hand, AI could undermine all previously difficult mathematical problems, it would threaten all modern cryptography. As a security officer, one should therefore use this transition moment as an opportunity: intensified AI-powered testing of new algorithms can lead to more robust cryptanalysis literature and increased confidence in the chosen procedures.
Source: simonwillison.net · Published 29 July 2026
Lumi AI News — AI-assisted curation in accordance with Article 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.3.