---
schema_version: '1.0'
id: news-20260730-4a3383
url: https://gotosocial.chinng-lab-srv.dev/aiganai-liang-zi-an-hao-hawknoqian-xian-wo60shi-jian-defa-jian-claude-mythosnochong-ji/
url_hash: 4a33833b4aeeddc2c94621378f04747503a38bb68833a12f1ace914047bdc74c
canonical_url: https://gotosocial.chinng-lab-srv.dev/aiganai-liang-zi-an-hao-hawknoqian-xian-wo60shi-jian-defa-jian-claude-mythosnochong-ji
source: ghost-chinng-lab
category: news/general
category_raw: 耐量子暗号
region: JP
tags:
- 耐量子暗号
- PQC
- HAWK
- Claude Mythos
- Anthropic
- NIST
- AES
- 暗号解析
lang: ja
published_at: '2026-07-30T11:34:24Z'
fetched_at: '2026-07-30T12:39:04.989333Z'
updated_at: '2026-07-30T12:39:23Z'
status: published
content_hash: null
content_changed_at: null
license_note: full
summary: Claude Mythos（Anthropic開発）が、耐量子暗号HAWK候補の欠陥を60時間で発見した。2年間の専門家査読をすり抜けた設計における非自明な自己同型を利用した攻撃により、HAWK-256の鍵強度が半減（2^64から2^38へ低下）。同時にAES-128の7ラウンド削減版への攻撃も自律的に開発し200～800倍の高速化を実現。実運用への影響はないが、AI検証速度が人間の査読を追い越す現象を象徴しており、暗号標準化プロセスの再設計と生成・検証の非対称性が焦点に。
summary_source: llm
summary_en: Claude Mythos (Development of Anthropic) discoveredHAntum Crypto HAWK
  candidate defects in 60 hours. HAWK-256’s key strength is reduced by half due to
  a non-self-identified attack in a 2-year expert-reviewed design. At the same time,
  AES-128’s 7-round reduction version is developed autonomously and speeds up from
  200 to 800. There is no influence on actual operation, but the AI verification rate
  symbolizes the phenomenon of overcoming human読er review, and the a対metry of redesign,
  generation and verification of the encryption standardization process is focused。
entities:
- name: PQC
  type: UNKNOWN
- name: HAWK
  type: concept
- name: Claude Mythos
  type: artifact
- name: Anthropic
  type: organization
- name: NIST
  type: organization
- name: AES
  type: artifact
- name: Claude
  type: person
- name: Mythos
  type: person
key_facts: []
related: []
related_auto:
- name: 量子鍵配送(QKD
  type: technology
  weight: 1.0
- name: Claude Mythos Preview
  type: event
  weight: 1.0
- name: Anthropic Frontiers Red Team
  type: UNKNOWN
  weight: 1.0
- name: Ars Technica
  type: organization
  weight: 1.0
- name: non-trivial automorphism
  type: concept
  weight: 1.0
title: AIが耐量子暗号HAWKの欠陥を60時間で発見——Claude Mythosの衝撃
---

# AIが耐量子暗号HAWKの欠陥を60時間で発見——Claude Mythosの衝撃

## TL;DR
Claude Mythos（Anthropic開発）が、耐量子暗号HAWK候補の欠陥を60時間で発見した。2年間の専門家査読をすり抜けた設計における非自明な自己同型を利用した攻撃により、HAWK-256の鍵強度が半減（2^64から2^38へ低下）。同時にAES-128の7ラウンド削減版への攻撃も自律的に開発し200～800倍の高速化を実現。実運用への影響はないが、AI検証速度が人間の査読を追い越す現象を象徴しており、暗号標準化プロセスの再設計と生成・検証の非対称性が焦点に。

## Key Points
- 耐量子暗号 / PQC / HAWK / Claude Mythos / Anthropic / NIST / AES / 暗号解析

## Details
(本文なし。リンク先参照)

## Source
元記事: [AIが耐量子暗号HAWKの欠陥を60時間で発見——Claude Mythosの衝撃](https://gotosocial.chinng-lab-srv.dev/aiganai-liang-zi-an-hao-hawknoqian-xian-wo60shi-jian-defa-jian-claude-mythosnochong-ji/) — published 2026-07-30T11:34:24Z