Discovering cryptographic weaknesses with Claude

Researchers at Anthropic employed the Claude Mythos tool to identify mathematical vulnerabilities in encryption algorithms, including HAWK and a variant of AES that is more susceptible to attack. Although the findings do not have a practical impact on current computer systems, the research process is intriguing. The researchers shared the prompts used to guide the tool, highlighting the importance of proper configuration in obtaining valuable results. The tool operated for 60 hours and required human intervention to prevent it from stalling and produce publishable results. The study also led to the development of a new evaluation framework, CryptanalysisBench, in collaboration with several universities. This news is significant as it demonstrates how artificial intelligence can be leveraged to enhance the security of computer systems, and how research in this area can yield valuable discoveries. The application of artificial intelligence in cybersecurity is a rapidly evolving field, and studies like this underscore the importance of continued exploration of its potential.

Read the original article on Simon Willison

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.