Anthropic's classifiers add 24% to compute costs — and AI refusal still fails
Anthropic says one classifier adds 24% to compute costs, yet jailbreakers cracked Fable 5 in under three days. An investigation examines how AI refusal works — and fails.
Topic
Topic
Anthropic says one classifier adds 24% to compute costs, yet jailbreakers cracked Fable 5 in under three days. An investigation examines how AI refusal works — and fails.