Home / Technology / AI Cracks Encryption: A Cybersecurity Wake-Up Call
AI Cracks Encryption: A Cybersecurity Wake-Up Call
28 Jul
Summary
- AI model Claude Mythos Preview found new attacks on crypto algorithms.
- Flaws identified in a weakened version of Advanced Encryption Standard.
- AI attacks were 200 to 1,000 times faster than previous human research.
An advanced AI model, Claude Mythos Preview, has identified new vulnerabilities in a weakened form of a widely used encryption standard. Researchers at Anthropic announced that their AI was able to detect novel attack vectors against cryptographic algorithms that secure sensitive online activities.
These algorithms protect everything from financial transactions to private communications. The flaws uncovered do not affect encryption standards currently in use, meaning immediate risks to modern systems are minimal. However, the research utilized a diluted version of the Advanced Encryption Standard (AES) to test AI capabilities.
In testing, Claude Mythos Preview demonstrated an ability to break this weakened AES version with assaults between 200 and 1,000 times faster than previously recorded human efforts. This rapid advancement by AI in a mathematically complex field like cryptography challenges previous assumptions about AI's limitations and signals potential future risks to foundational internet security.
