Anthropic’s Mythos Finds Weaknesses in HAWK and a Weaker AES Variant
Summary
Anthropic says its Mythos model found ways to weaken HAWK and a weaker AES variant in a week. The article discusses the potential benefits and risks of using frontier AI to test encryption algorithms.
Key points
- Anthropic reports that Mythos found ways to reduce the robustness of HAWK and a weaker version of AES.
- The work took about a week, with small teams guiding Mythos’s mostly autonomous research.
- The article says the findings do not affect current computer systems and do not amount to fully breaking either algorithm.
- HAWK was submitted as a potential signature algorithm in 2023; AES was adopted in 2001.
- The author argues AI-assisted testing could help identify weaknesses before attackers do, but could also increase risks if attackers find exploitable weaknesses first.
Article Details
- Topic
- Anthropic’s Mythos model finding weaknesses in HAWK and AES encryption algorithms