Anthropic has updated the biology safeguards on its Claude Fable 5 artificial intelligence model, allowing it to handle a wider range of biology-related requests while reducing instances in which users are moved to another model.
The changes cut biology-related fallbacks by about 85% across Anthropic's products during testing, the company said on Friday. A fallback occurs when the system moves a query from Fable 5 to another model after its safeguards flag a biology-related request.
The update expands the range of biology tasks that Fable 5 can handle, including everyday health and education queries, interpretation of laboratory results and information about symptoms. Anthropic said healthcare professionals could also receive greater assistance with clinical tasks.
"We're making updates to Claude Fable 5's biology safeguards in a way that substantially reduces false positives," Anthropic said in a blog post. "Fable 5 users will now experience many fewer 'fallbacks'."
Why Anthropic Changed Safeguards
Anthropic had introduced Fable 5 with broad biology classifiers because biological research can have both beneficial and harmful applications.
Scientific knowledge used for areas such as disease treatment could also be misused in the development of biological weapons, the company said. Anthropic's capability assessments found that Fable 5 "could provide significant uplift" to a malicious actor by providing capabilities that are not easily available elsewhere.
The company said it accepted tighter restrictions when it launched the model because "the cost of Fable being misused in a dual-use domain like biology could potentially be catastrophic."
Anthropic has now refined those safeguards to reduce the number of legitimate queries that get blocked.
"We believe the greatest opportunity for AI to positively affect the world is in biology and medicine, and we're investing significantly in building a responsible way to give biologists frontier access," the company said.
ALSO READ: OpenAI, Anthropic Model Tests Reveal More Unsanctioned' Actions
Limits Remain
The update does not give Fable 5 unrestricted access to professional biology research. Queries that Anthropic considers dual-use, including those involving virology, toxicology and molecular design, will continue to trigger a fallback to its Opus 5 model.
That means Fable 5 remains unavailable for some professional biology research and drug development tasks.
Anthropic said it remains "committed to closing that gap through trusted access pathways for frontier biology capabilities".
The company spent several weeks rewriting the rules underpinning its safety classifier. It incorporated feedback from internal and external experts before retraining and testing the system to check that it continued to identify harmful content while allowing more benign and educational requests.
Anthropic said the revised system would not eliminate all incorrect classifications.
"There's still much more to be done to refine our safeguards," the company said. "There will inevitably remain false positives - requests that fall within the classifier's safety margin where the request is very low-risk but where the classifier still fires."
Anthropic said it is "fully committed to developing a safe, scalable path for researchers to use our most capable models via trusted access pathways."
ALSO READ: Anthropic To Design Custom Silicon For Claude; Continues Nvidia, Google, AWS Tie-Ups
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.