BigHat puts AI drug hype back in the lab, where antibodies still have to prove themselves

In an interview published by STAT News, the CEO of BigHat Biosciences draws a useful boundary around artificial intelligence in biotech: AI can help design therapeutic antibodies, but it does not erase biology, regulation, or clinical risk.

That tonal correction matters. Over the past few years, AI in drug development has often been sold as an industrial shortcut: fewer attempts, faster candidates, a shorter road to therapy. BigHat is interesting because it works in a narrower and more concrete domain. The San Carlos company applies machine learning to antibody therapy design, a problem where computational prediction can be tied to measurements of biological properties in candidate molecules.

Even there, the distance between finding a better candidate and proving a medicine remains large. A model can help search design space, filter variants, and steer experimental work. It cannot by itself establish safety, dosing, durability of effect, rare adverse events, or how a therapy behaves in real patients. That is why the most useful part of this story is the refusal of the simple formula in which every model improvement automatically becomes a medical breakthrough.

For anyone following medical technology, BigHat’s example is a good maturity test for the broader AI-biotech narrative. If AI is described as a replacement for pharmaceutical development, the claim is too weak. If it is described as a tool that can improve early decisions, reduce some blind search, and support better experimental iteration, it becomes more serious.

The regulatory context points in the same direction. The U.S. FDA is already discussing AI and machine learning in drug development, but regulators are not interested only in elegant models. They need to understand how a model was used, what it measured, where it can fail, and how results are validated outside the computational environment.

So this is not a story about one sweeping breakthrough. It is more useful as a signal that serious AI-biotech operators are starting to speak with more precision. In antibody development, AI can be a powerful instrument, especially when it accelerates iteration between design and laboratory testing. But medical value emerges only when a candidate survives experimental validation, manufacturing constraints, and clinical reality.

The short version is this: less mythology, more engineering discipline. BigHat’s message in the STAT interview is not that AI will invent medicines on its own, but that it may make some early decisions smarter. In an industry where failures are expensive and patients are treated by therapies rather than slide decks, that is already a substantial claim without extra inflation.