A research team from the Universitat Rovira i Virgili (URV) has developed CoCoGraph, a new artificial intelligence tool capable of generating millions of chemically valid new molecules. The findings were recently published in the journal Nature Machine Intelligence. This system represents a significant advancement in computational chemistry, potentially accelerating the discovery of new drugs and sustainable materials. Unlike text or image generators such as ChatGPT or Dall-E, CoCoGraph operates on the rules of chemistry rather than natural language or visual aesthetics. It utilizes a diffusion model, a technique adapted from image generation, to create molecular structures. The process involves artificially breaking the bonds of a real molecule to create a disordered state, then training the AI to reverse this process and reconstruct coherent, valid structures. Because molecules are discrete rather than continuous like images, this task presents unique mathematical challenges. The core innovation of CoCoGraph is its ability to ensure 100% validity in its outputs. The model directly incorporates fundamental chemical laws, guaranteeing that every atom maintains the correct number of bonds. Consequently, it avoids producing impossible structures common in other generative models. Additionally, the system is more efficient, requiring fewer parameters and less computing power while generating molecules faster than competitors. The scale of this achievement is immense. While the number of known molecules is limited, scientists estimate there could be up to 10^60 possible molecular combinations. Current AI tools often struggle to navigate this vast chemical space effectively. In comparative tests against state-of-the-art models, CoCoGraph produced molecules that were chemically more realistic than those from other systems for approximately two-thirds of 36 analyzed physicochemical properties, including solubility and structural complexity. To validate the realism of the generated molecules, the team conducted an experiment with 121 chemistry experts. Participants were shown pairs of molecules, one real and one AI-generated, and asked to identify the real one. Experts failed to distinguish between the two in roughly 40% of the cases, indicating that the AI creates highly convincing chemical structures. Although the system cannot yet be prompted to design molecules with specific functions, it has successfully identified variants similar to paracetamol and explored techniques to slightly modify existing molecules. Researchers emphasize that this development is a foundational step toward a future where AI can design bespoke molecules. The team aims to eventually allow users to specify precise requirements, such as solubility, non-toxicity, or specific applications. Success in this area could transform pharmacology, chemistry, and materials science by unlocking a vast, previously unexplored universe of chemical solutions.