Evan Travers retweetledi

So apparently Anthropic left 3,000 internal docs in a public data lake and security researchers found them.
The docs reportedly describe a new model called claude mythos that sits above opus in a new tier called capybara. Their own draft allegedly says it's "far ahead of any other AI model in cyber capabilities".
If true, the company building the most security-sensitive AI on earth got popped by a config flag.
English






























