Two of the world's most powerful AI companies are betting that their next big market isn't faster coding or better chatbots, but cures for diseases. Anthropic launched Claude Science on June 30, a specialized AI workbench designed to help scientists discover new drugs and accelerate research, while OpenAI released GPT-Rosalind in April for the same purpose. Both moves signal a fundamental shift in how AI companies are positioning themselves, moving beyond consumer tools into the infrastructure of scientific discovery itself.

Why Are AI Companies Suddenly Focused on Drug Discovery?

The timing matters. Anthropic has spent recent weeks navigating government scrutiny over its most powerful models, Fable 5 and Mythos 5, with concerns about their capabilities and potential risks. Claude Science represents a strategic pivot: instead of defending why powerful AI exists, Anthropic is now arguing that powerful AI is necessary because it can help cure diseases. The company is not simply selling a tool; it is positioning itself as an AI-native drug developer by launching its own internal preclinical drug-discovery programs focused on neglected and rare diseases.

This is a calculated reframing. A chatbot that answers questions is one thing. A system that helps identify new drug candidates for rare genetic diseases is something society might accept more risk for. The moral equation shifts when the potential benefit is medicine rather than convenience.

What Can Claude Science Actually Do?

Claude Science integrates more than 60 scientific databases and connects tools like PubMed, Jupyter, and R, with pre-built capabilities for genomics, proteomics, structural biology, and cheminformatics. Every result traces back to the exact code and environment that produced it, which matters for reproducibility in science. At the launch, Alexander Tarashansky, who led its development, demonstrated the system autonomously identifying new drug candidates for phenylketonuria, a rare genetic disease.

The early internal results are striking. Using Anthropic's Mythos 5 model, researchers reported accelerating parts of the drug-discovery process by roughly ten times, and the model generated novel molecular-biology hypotheses and ran largely autonomous genomics research. The company's models already demonstrated strong biological reasoning by completing a complex gene-therapy task involving adeno-associated viruses, a capability that could accelerate drug development.

However, Anthropic's own leaders acknowledged the cold reality: Claude may speed up research and help scientists find better starting points, but it does not get to skip biology. A molecule that looks good in a model still has to survive lab testing, show safety and efficacy, pass animal studies, enter human trials, clear regulators, and be manufactured and delivered.

How Are AI Companies Approaching Safety in Drug Discovery?

The dual-use problem is real. A model that helps scientists understand biology could help medicine, but it could also help someone misuse biological knowledge. Anthropic's approach involves layered safeguards. Fable 5's classifiers reroute biology and chemistry queries to the weaker Opus 4.8 model, and Anthropic deliberately made these safeguards overly conservative, blocking most queries tied to biology work even at the cost of catching harmless ones.

• Capability Access as a Gate: Sensitive biological workflows are placed behind tighter controls, with basic research help available to many users but advanced capabilities restricted to vetted researchers.
• Deliberate Over-Caution: Anthropic's safeguards are intentionally conservative, blocking more queries than strictly necessary to ensure harmful knowledge is not leaked.
• Dual-Use Acknowledgment: The company explicitly recognizes that improving a model's ability to design complex therapeutic molecules also develops dangerous knowledge, requiring equal investment in safeguards and capabilities.

"A model that gets better at designing complex therapeutic molecules is also a model that develops dangerous knowledge, and that the company needs to put as much work into safeguards as into capabilities," stated Eric Kauderer-Abrams, Anthropic's life sciences head.

Eric Kauderer-Abrams, Life Sciences Head at Anthropic

Why Target Rare Diseases First?

Anthropic's focus on neglected and rare diseases is a deliberate strategic choice. The logic is interesting: with many rare diseases, the biology is often clear because a single damaged gene is the cause, while the economics of developing a treatment are challenging. Clear biology, weak market. That is the gap Anthropic says it wants to target. Traditional drugmakers find these diseases unattractive because the patient populations are small and the profit potential is limited, but AI can help solve the research problem without needing a blockbuster market.

The company has not named the specific diseases it is pursuing, and stressed the work is preclinical, the stage before human testing, with no decision yet on whether it would bring candidates to market, license them, or hand them to partners.

How Does This Compare to OpenAI's Approach?

OpenAI released GPT-Rosalind in April, an AI model built to speed up research and drug discovery, making Claude Science a direct competitive response. Both companies are racing to establish themselves as essential infrastructure for scientific discovery, but they are taking slightly different paths. Anthropic is going further by actually conducting its own drug-discovery research, not just providing tools.

This mirrors the template set earlier by Google DeepMind with AlphaFold, which changed how scientists think about protein-structure prediction and gave AI one of its cleanest scientific wins. Anthropic has a credibility advantage here: CEO Dario Amodei is a PhD scientist, and earlier this month the company hired John Jumper, the Nobel laureate who led AlphaFold, away from DeepMind.

What Does This Mean for the Broader AI Industry?

The move signals that frontier AI companies are no longer content to be software vendors. They want to be infrastructure for entire industries. Claude Science is elevated to the same rank as Claude Code and Claude Cowork, a signal that Anthropic is treating AI's scientific applications as a core product rather than a side experiment. The deeper play is obvious: Anthropic does not want Claude to be a clever assistant; it wants Claude to become infrastructure for science.

This is also a business strategy. Claude Science is not charity. It is enterprise strategy designed to build long-term relationships with pharmaceutical companies, research institutions, and biotech startups. By positioning itself as essential to drug discovery, Anthropic creates a defensible market position that is harder to disrupt than consumer chatbots.

The risk is that AI hype in drug discovery could outpace reality. Many companies are making claims they cannot back up, and the technology is not a panacea. But if Claude Science or GPT-Rosalind can genuinely accelerate the discovery of treatments for neglected diseases, the public debate about powerful AI shifts from restriction to necessity.