AI
Anthropic's Claude Mythos Preview cracks two cryptographic algorithms in under a week
Anthropic announced that Claude Mythos Preview, working largely autonomously, discovered previously unknown weaknesses in two well-studied cryptographic schemes. Within 60 hours it found an attack that cut the key strength of HAWK — a post-quantum digital signature algorithm that had survived two years of expert review — by half. Separately, it found a way to speed up an attack on a reduced version of AES by 200 to 800 times. Each result cost roughly $100,000 in API usage, and Anthropic disclosed the findings in advance to the algorithms' authors and to US government and industry partners.
Anthropic was careful to note that neither result breaks systems in production today: HAWK has not been deployed, and the AES attack targeted a weaker variant rather than the full cipher. The lab framed the work as a proof of concept for defensive applications — using frontier AI to stress-test algorithms before they are widely adopted — and released two technical papers alongside CryptanalysisBench, a new benchmark developed with academics at ETH Zurich, Tel Aviv University, and the University of Haifa to study LLMs' cryptanalysis abilities.
OpenAI publishes eight case studies on coding agents accelerating scientific research
OpenAI released a report examining how coding agents are being used across eight case studies spanning industry and academic research labs, covering tasks from routine code maintenance and targeted optimization to complete system redesigns. The report argues that while agents can reliably execute on ambitious engineering projects, human researchers must still define the goals, verify results, and handle long-term stewardship — framing AI not as a replacement for scientists but as a tool that frees them to focus on higher-level research direction.
This briefing is part of the archive.
Get the app for full archive access — every daily briefing, sourced and attributed.