Developing a competing AI model with Claude is already against their TOS, and there was a period where they silently degraded quality if they thought that's how you are using Fable. People were not amused and presumably Anthropic currently does not sabotage your attempts, apart from regular TOS enforcement
replyCan this company ever be trusted?
replyClaude Fable 5's model card disclosed safeguards that would invisibly degrade output for requests linked to frontier LLM development. The research community called it "sabotage". Dean Ball: "degrading performance on ML research without telling the user is shockingly hostile" — including because it would corrupt third-party safety evaluators of Anthropic's own models.
They already put limits on machine learning research, with explicit intent to sabotage competitors
replyLucky we have Kimi Pro then
reply