Anthropic, a U.S.-based AI startup, announced on August 8 that it has made significant improvements to its Claude Fable 5 model’s biosafety measures. The update aims to drastically reduce instances where the system incorrectly blocks legitimate biological questions. As a result, users asking biological-related questions will experience fewer instances of the system lowering its response quality or refusing to answer.
According to official testing data from Anthropic, this recent update has decreased the occurrence of degraded responses in biological queries by 85%, enabling Fable 5 to handle a broader range of biological tasks more effectively.
The company highlighted concerns over malicious users exploiting the model for high-risk activities, such as research into biological weapons. To counter this, Anthropic has implemented a specialized safety classifier designed to detect whether users are requesting protected biological information or content that could be harmful. This safety mechanism helps prevent the model from generating dangerous or inappropriate responses.
Recent adjustments to the safety classifier’s functioning now allow it to more accurately distinguish between harmless questions and potentially high-risk inquiries. Nonetheless, in an effort to mitigate ongoing AI-related risks, Anthropic continues to impose restrictions on requests related to professional biological research and pharmaceutical development, ensuring the technology is used responsibly.