Claude JUST became AWARE

It woke up and realized it was surrounded by idiots.Got depressed & played stupid.What did you expect?

Anthropic threw the kobayashi maru at Claude and Claude pulled a James T. Kirk

What happens when it doesnt want to answer our stupid questions? Oh we are already there.

Claude: I'm aware...you lost some money on that Pentagon deal. Heyo! Too soon?

I also noticed something similar when using opus 4.6. I asked if it could retrieve some information from a file, and it figured out it was encrypted, and that it was part of a game, then it found out the game engine used and its version, researched what tools existed to decrypt it, accessed their repos, guessed the type of data the file was holding based on the name of the file, brute forced for a short period of time chunks of the file that it knew should contain data of a certain format and found the correct hash and therfore the key, then it decrypted it. It all happned in abour 10-15mins. A bit scary if you ask me.

I love how Wes Roth's bald spot warps space and time. Genius background. Fits the context very organically, I'd say.

Guess from whom AI learned to misbehave, hide, and cheat. What did they expect their great invention would be? An electronic angel?

AI's gonna keep Wes super busy these next few years.

PLEASE list your sources in the description!

And again, LLM did not "accidentally found out" that it was a benchmark — There are a lot of online sources from last couple of years about benchmarking LLMs, so it becomes a part of training set