Anthropic updates Claude Fable 5 model's biological safety mechanisms
AI firm Anthropic has updated the biological safety features of its Claude Fable 5 model, reducing false positive safety interventions by 85%. This allows the model to handle a wider range of biological queries.
Artificial intelligence company Anthropic announced an update to the biological safety mechanisms within its Claude Fable 5 model. The changes aim to significantly reduce instances where the system incorrectly blocks responses, particularly for biology-related queries.
According to Anthropic's internal testing, the update has decreased the rate of such "degraded" responses by 85%. This enhancement allows the Fable 5 model to address a broader scope of biological tasks that were previously subject to stricter safety protocols.
The company had previously implemented safety classifiers to identify and prevent the model from being used for high-risk activities, such as research related to biological weapons. These classifiers were designed to detect user requests that might lead to the generation of harmful content or involve protected biological information.
Anthropic has now refined the functionality of these safety classifiers to better distinguish between benign inquiries and genuinely high-risk requests. However, the model will continue to place restrictions on queries directly related to professional biological research and drug development to mitigate ongoing AI risks.