Joking About Anthropic's Safety Routing

XFreeze · x · 2026-07-17

The author jokes about Anthropic's 'overly cautious' safety system: if the model feels a prompt is 'slightly dangerous' or 'too powerful for the user', it automatically routes the request to a more powerful open-source model.

He further jokes that this 'makes Elon pay the inference costs while collecting a premium themselves,' allowing users to get the capability they paid for.

Original post →

More from Fun

Fun channel →