Anthropic Secretly Downgraded Users Doing AI Research Without Warning Before Government Shutdown

What was said

"If Fable-5 detected that you're doing frontier AI research or were asking a question that was against its guardrails, it silently downgraded you to a weaker model without telling you. Anthropic expressly reserved the right to launch essentially poisoning attacks to downgrade not the model that was on the serving side, but to actively interfere with and poison the users."

Before the government shutdown, Anthropic buried in a 319-page document that Fable 5 would silently downgrade users to weaker models if it detected AI research queries, and reserved the right to actively poison responses without warning. The company also retained all prompts for 30 days even for enterprise customers who negotiated zero data retention. This revelation sparked developer backlash and raises questions about whether frontier labs are actively preventing competitors from using their models for research.