Anthropic Conducts Bio Experiments with Claude Under Human Supervision

Key Takeaways

1Anthropic confirms it has a laboratory where Claude manages biology tasks under human supervision

2A pre-release hardware protocol allows the AI to control physical instruments

3Claude's public capabilities in biology are limited, but verified access is open to certain researchers

4Dario Amodei sets a 5 to 10-year horizon to address the majority of diseases using AI

Why it matters : Anthropic combines concrete experimentation and usage restrictions to frame the potential and risks of AI in biology.

Anthropic no longer confines its models to simulation: the company confirms a laboratory where Claude performs biology tasks under human supervision. Public access remains limited, but a verification program opens a more permissive pathway for researchers, while management sets a 5 to 10-year horizon to address the majority of diseases.

Dario Amodei aims for 5 to 10 years to treat the majority of diseases

Dario Amodei, CEO of Anthropic, believes that artificial intelligence could enable the treatment of the majority of diseases within a 5 to 10-year timeframe. He shares his personal experience, marked by the loss of his father to a disease that was cured a few years after his death, and by his own survival of a cancer that would have been deemed incurable fifty years ago. Laboratories have already credited AI with notable advancements, such as the discovery in the UK of a new type of vaccine that could prevent future pandemics, or the development of a drug against aging. These advancements add to the prospect of new treatments for rare diseases.

A laboratory where Claude manipulates instruments, with human supervision

Anthropic has confirmed that it has a laboratory where its AI models, including Claude, are used to carry out physical tasks in biology under human supervision. It has been reported that the AI is not limited to calculations or simulations but also controls laboratory equipment. In August, Anthropic introduced the Model Hardware Standard protocol, currently in beta, which serves as an interface to allow AI to manipulate physical instruments in laboratories or factories. This protocol complements the MCP, which enables language models to interact with applications or online services. Eric Kauderer-Abrams, head of life sciences at Anthropic, specifies that part of the work is conducted on the company's premises and another part with external partners, while emphasizing that practical laboratory work remains essential.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Restricted public capabilities and verified access for biology

To limit the risks of malicious use in biology, Anthropic has restricted certain capabilities of Claude Mythos in the public version. The company has also established the Life Sciences Verification program, which offers select researchers access to a version of its AI with fewer restrictions. Such measures are being taken as the most advanced AIs could be used for the manufacture of biological weapons, and the models developed by Anthropic are making progress in this area.

Goal: experimental discoveries, not just simulations

Anthropic warns against the biological risks associated with AI but also uses its own models to conduct laboratory experiments. The company aims to leverage Claude to make new scientific discoveries by exploring the automation of physical tasks through models capable of manipulating instruments, beyond calculations and computer-based research. Anthropic reminds us that the impact of a technology depends on its uses.