Artificial intelligence has moved from a promising investment thesis to a clinical reality in drug discovery. At the BIO International Convention in San Diego last week, the industry celebrated a historic milestone: Insilico Medicine's Rentosertib, a drug discovered entirely by AI, completed Phase IIa trials and showed meaningful improvement in patients with idiopathic pulmonary fibrosis. This marks the first peer-reviewed evidence that generative AI can identify both the disease target and design a working compound without human chemists.

What Does It Mean That an AI Drug Actually Works?

The Rentosertib trial enrolled 71 patients in a double-blind, placebo-controlled study. Those who received 60 milligrams once daily showed a mean improvement in lung function of 98.4 milliliters, while the placebo group experienced a decline of 20.3 milliliters. The results were published in Nature Medicine on June 3, 2025.

What makes this different from earlier AI drug discoveries is the pipeline itself. Previous AI tools helped human chemists screen existing compounds or optimize molecules they had already identified. Rentosertib represents something categorically different: the AI analyzed multi-omics data to propose the TNIK kinase target, then designed the compound by navigating chemical space through reinforcement learning. No human chemist made the key decisions. The AI did.

The underlying technology comes from RFdiffusion, a protein design method developed by Nobel laureate David Baker's team at the University of Washington's Institute for Protein Design. The approach applies the same denoising technique used in AI image generation to protein structures. During training, known protein structures are progressively corrupted with random noise over roughly 200 steps, and the network learns to reverse that process. At inference, the model starts from purely random atomic frames and converges on a coherent protein backbone.

The practical impact is dramatic. Earlier design methods required testing tens of thousands of candidate molecules before finding one that worked. RFdiffusion reduced that number to as few as one per design challenge in laboratory experiments. A December 2025 update, RFdiffusion3, extended the approach to atom-level design across all biomolecule types, including DNA and small-molecule interactions, and was released as open-source software.

Why Is the Biotech Industry Suddenly Asking Different Questions?

The tone at BIO 2026 shifted noticeably. For the past three years, investors asked whether AI drug design could work at all. Now they are asking what is working, what is not, and how fast they need to move. More than $11 billion flowed into AI and machine learning drug discovery and licensing in 2025 across 348 funding rounds, according to DealForma. That volume of capital now demands accountability, and the Rentosertib result provided it.

Generative protein platforms including Xaira Therapeutics, Generate Biomedicines, and Isomorphic Labs have moved from filling pipelines with candidates to producing clinical-stage molecules ready for human testing. Xaira Therapeutics, co-founded by David Baker and launched in April 2024 with more than $1 billion in committed funding, builds its platform directly on RFdiffusion's architecture.

How to Understand the Competitive Landscape Reshaping Drug Discovery

• Chinese Biotech Originators: Chinese companies have transitioned from contract manufacturers and fast-followers to world-class originators of pharmaceutical intellectual property. In the 2024-2025 deal cycle, Chinese biopharma companies secured more than $30 billion in oncology licensing deals alone, according to industry analyses.
• Next-Generation Technologies: Chinese antibody-drug conjugate (ADC) innovators including Kelun-Biotech, Argo Biopharmaceutical, and MediLink have developed linker technologies that address a structural limitation of first-generation Western ADCs: premature release of toxic payloads into the bloodstream before reaching tumor cells.
• Global Licensing Momentum: Merck, Novartis, and GSK signed deals totaling over $10 billion for Chinese cardiovascular and oncology assets in late 2025 alone, indicating that Chinese science is now competing on scientific merit, not just manufacturing cost.
• GLP-1 Competition: With the world's largest population of diabetic and overweight patients, Chinese companies built a domestic pipeline of GLP-1 receptor agonists, triple agonists, and oral small-molecule alternatives to injectable peptides that are now competing directly against first-generation Western products in global licensing markets.

The BIOSECURE Act, signed into law on December 18, 2025, as Section 851 of the National Defense Authorization Act for Fiscal Year 2026, prohibits U.S. federal agencies from contracting with entities that use biotechnology equipment or services from designated Chinese companies of concern. The law has already shifted how U.S. biotechs structure their contract manufacturing and research relationships, with large Chinese contract development and manufacturing organizations linked to the government seeing U.S. clients diversify their supplier bases in anticipation of enforcement provisions expected around 2028.

However, the BIOSECURE Act operates at the manufacturing and service layer. It was designed for a competitive landscape in which China's advantage was cost-efficient execution. That landscape has fundamentally changed. The law cannot reach the science layer itself. Chinese biotech companies are now originating superior molecules, not just manufacturing them cheaper.

For investors and executives at BIO 2026, this distinction was the most important strategic fact of the convention. The United States can restrict manufacturing partnerships, but it cannot legislate away Chinese scientific innovation. The AI drug discovery milestone that should have been a moment of American triumph instead highlighted a competitive reality: the technology works, capital is flowing, and the race is already underway globally.