Anthropic's Claude Mythos has just found a way to break post-quantum encryption.

Claude Mythos is Anthropic's most advanced AI model, and it cracked two different encryption systems.

The first is a newer type of encryption designed to remain secure even in the face of future quantum computers.

Claude discovered a flaw that cut its security in half, meaning it is now twice as easy to crack as experts had imagined.

The second is an encryption standard used to protect most daily internet traffic, such as your emails, bank transactions, and browsing activity.

When researchers first asked Claude to attack the second system, it refused, stating that the task was essentially impossible and that no new findings could be made.

The researchers pushed back and told it to keep trying. Over the next three days, working almost entirely autonomously, Claude generated hundreds of millions of words of analysis and ultimately discovered the winning method on its own.

Both discoveries were made on test versions specifically built to study vulnerabilities, rather than on real-world systems actually protecting your data today.

Therefore, nothing you are currently using is at risk.

The computational cost for each of Anthropic's discoveries was approximately $100,000.