Anthropic said on Tuesday that its Claude AI found new weak spots in two encryption systems. Encryption is the math that hides your messages, logins and bank details from strangers.
Nothing you use today is at risk. But the news matters for a different reason. A machine did work that top human experts had missed.
What Is HAWK, and What Broke
HAWK is a digital signature. Think of it as an ID check for websites and messages. It was built to survive future quantum computers. Experts reviewed it for two years and found nothing. Claude found a hidden pattern in HAWK’s math in 60 hours.
Follow us on X to get the latest news as it happens
By Anthropic’s own figures, cracking the smallest HAWK key fell from 2^64 steps to 2^38. That is roughly 67 million times less work. The method is in Anthropic’s paper.
The fix is simple but painful. Double the key size. That removes the main reason to pick HAWK.
The AES Result, Explained Simply
AES (Advanced Encryption Standard) scrambles most internet traffic. It has guarded data since 2001.
Anthropic did not break AES. It attacked a weaker version with 7 of the usual 10 scrambling rounds.
Claude invented a shortcut it named the Möbius Bridge. The trick removed one guessing step.
That made the best known attack 200 to 800 times faster.
Claude worked alone for three days. Two humans then spent nearly a month checking its answer.
Why This Matters
Checking the work is now the slow part. Anthropic says humans cannot keep pace with the model.
Code designs have failed late before. In 2022, researchers cracked SIKE, another candidate, in about an hour on a laptop.
Bitcoin (BTC) and Ethereum (ETH) rely on a different system. Experts still argue over when quantum computers break it. These attacks do not touch it.
The clock is running anyway. Google plans to switch by 2029 under its migration timeline. US agencies face a 2031 quantum deadline.
What to Watch Next
NIST (National Institute of Standards and Technology) must now weigh a public attack on a scheme it was still considering. Anthropic hinted at more results it has not released yet. It also plans a workshop with academics in the coming weeks.
Would you trust an AI to test the locks on your own data?
Bitrue