Anthropic has released Claude Fable 5, billed as a powerful tool for biology and beyond. Yet, it struggles with basic queries on cell membranes or mRNA vaccines, deferring to older models instead.

The restrictions are due to fears over bioweapons; Anthropic claims these ‘overly conservative’ safeguards aim to prevent malicious use of AI in biological research. But the limitations raise questions about the model’s true abilities and potential.

While Claude Fable can provide answers on cybersecurity or even give a basic overview of TNT, it struggles with prompts related to biology work, such as prions or anthrax. These restrictions are deliberate — Anthropic is working hard to improve its detection but at the cost of usability.

The tradeoff between safety and utility highlights an ongoing tension in AI development: how much can we trust these powerful tools without compromising security? As AI models like Fable continue to evolve, so too will the balance between progress and precaution.