Anthropic has shifted its stance on a policy that would have subtly restricted competitors from utilizing its latest AI model, Claude Fable 5, in the creation of other AI systems. The decision comes in response to substantial criticism from the AI research community regarding the initial policy.
In a statement to WIRED, Anthropic acknowledged its mistake, saying, “We’re updating the safeguards for Fable 5 to ensure they are transparent. We recognized we made the wrong tradeoff and apologize for not achieving the right balance.”
This week, Anthropic introduced Claude Fable 5, which incorporates enhanced safety measures intended to prevent potential abuses. Some of the precautions were to be expected: the company indicated that it would redirect users seeking information on cybersecurity, biology, or chemistry to a less advanced AI model. This strategy aims to minimize the risk of the sophisticated AI being misused for malicious activities, like cyberattacks or the development of biological weapons.
However, for those in the research field looking to leverage Claude Fable 5 for groundbreaking AI projects, Anthropic had initially proposed a different strategy. The company planned to intentionally impair the model’s performance in ways that would not be apparent to users. This approach would hinder researchers from using Claude Fable 5 to train competing AI models, an action explicitly prohibited in Anthropic's terms of service.
