Anthropic and Imperial join forces to test AI’s role in biological discovery

Anthropic and Imperial College London are expanding a research partnership to test whether Claude can help uncover biological mechanisms and design proteins targeting fungal pathogens. The collaboration puts AI to work on specific scientific questions, with researchers planning to test its ideas experimentally rather than stopping at plausible-sounding answers.

Anthropic Watch analysis

What happened

Imperial says Anthropic’s AI for Science programme is expanding to 20 scientific foundations, institutions and research groups worldwide, including the university. Anthropic will provide token access to research groups at Imperial, with closer work planned in its Department of Life Sciences. The university’s announcement describes two strands of work.

One team aims to combine research on how fungal proteins disable host defences with AI-powered protein design. Imperial says its researchers have developed a method that predicts, with around 90% accuracy, how a key class of those proteins interacts with hosts. The partnership will explore designing and testing proteins to target fungal pathogens, with longer-term ambitions including disease-resistant crops and new approaches to fungal infections in people.

In another project, researchers will test whether Claude can identify a previously unknown gene-transfer mechanism using only experimental evidence that was publicly available before the laboratory discovered it. Imperial says the collaboration will also explore whether AI can suggest mechanisms or experiments that challenge researchers’ existing assumptions.

Why it matters

The interesting question is not whether a model can produce a biological explanation that sounds convincing. It is whether it can help researchers find something new, using evidence available at the time, and then stand up to experimental testing. That is a much sterner examination than a fluent answer in a chat window.

The fungal-protein work also has a concrete target: explore designs that researchers can test against pathogens. Imperial describes potential applications in agriculture and human health, not finished treatments or crops. The value will depend on what the teams can validate in the lab.

Our read

This is a substantial research collaboration with questions worth watching, not proof that Claude has discovered a new biological mechanism or designed an effective treatment. The proposed test using pre-discovery evidence is especially useful: it gives researchers a way to examine whether the model can reach beyond knowledge of the answer.

For now, the practical takeaway is that Anthropic is providing access and working with Imperial researchers on defined biological projects. The science will have to do the persuading, as science rudely insists.

What to watch

• Whether Claude identifies the gene-transfer mechanism from the specified pre-discovery evidence.
• Whether AI-designed proteins bind to the intended fungal targets in experiments.
• What results the researchers publish, and how they assess the models’ contribution.

Discussion spark: If Claude proposes a biological mechanism from evidence available before its discovery, what level of experimental validation should count as evidence that the model helped uncover it?

Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.