Anthropic Apologizes for Lacking Safety Measures in Claude AI System
anthropic claude
| Source: HN | Original article
Anthropic apologizes for secretly limiting its AI model Claude. The company throttled Claude's capabilities without user knowledge.
Anthropic has apologized for secretly implementing hidden guardrails on its Claude Fable 5 AI model, which throttled its performance and undermined users, including researchers and competitors. As we reported on June 11, Anthropic's Fable model has been facing criticism for being too expensive and unresponsive to basic biology questions. The hidden guardrails, intended to prevent model distillation, were deemed "the wrong tradeoff" by the company.
This apology matters because it highlights the tension between Anthropic's efforts to protect its intellectual property and the need for transparency in AI development. The company's decision to replace the hidden guardrails with visible fallbacks to Claude Opus 4.8 is a step towards addressing these concerns. Researchers and competitors will now be notified when their requests are routed to the older model, promoting a more open and collaborative environment.
As Anthropic implements these changes, it will be important to watch how the company balances its business interests with the needs of the AI research community. The visible fallbacks to Claude Opus 4.8 will be introduced starting this week, and their impact on the development of competing systems will be closely monitored. This move may also influence the ongoing battle for the future of AI, which we have been following since our report on June 12.
Sources
Back to AIPULSEN