Scientists report they are nearing a turning point in drug development as AI-designed antibodies begin to show properties comparable to commercial medicines. The progress comes just one year after biologists first used artificial intelligence to generate antibodies from scratch — an early demonstration that lacked potency and key pharmaceutical features. Now, multiple research groups and companies say their latest AI-generated molecules exhibit strong binding, stability and manufacturability, traits essential for therapeutic use.
Researchers told scientific outlets that these developments mark a major step toward democratizing antibody engineering, enabling smaller labs and start-ups to design sophisticated molecules without vast screening platforms. Scientists say the new tools dramatically increase precision by allowing researchers to define the exact molecular target and obtain AI-generated antibody designs that match those specifications at an atomic level. Traditional screening techniques, by contrast, often yield weak binders or antibodies that attach to the wrong region of a disease-related protein.
A series of breakthroughs over the past year has helped overcome a long-standing challenge: accurately modelling the flexible loop regions antibodies use to latch onto their targets. Updated versions of AlphaFold and new open-source models now predict these loops more reliably, driving rapid advances in design quality. Teams at MIT, Stanford and the Arc Institute recently reported AI tools that can design nanobodies — compact antibody-like molecules — with high success rates against targets linked to cancer, infectious diseases and other conditions.
At the same time, several companies have announced what they describe as the boldest progress yet: the creation of full-length, drug-like antibodies designed entirely with AI. Groups at Nabla Bio and Chai Discovery say their molecules show potencies similar to existing antibody drugs and can bind difficult targets, including GPCRs, which have long resisted conventional antibody design. Early lab tests also indicate favourable pharmaceutical properties, such as reliable manufacturing yields and tight specificity for intended targets. Some academic researchers have urged the companies to release full data so their claims can be independently evaluated.
Despite the optimism, experts caution that AI models still struggle to predict certain key features, including binding strength and the risk that completely new antibodies could trigger immune reactions in humans. For now, scientists expect several more years of refinement before AI-designed antibodies can be relied upon exclusively in therapeutic development. Still, clinical testing has already begun in adjacent areas. Generate Biomedicine launched a large trial of an antibody for severe asthma that was improved using AI to enhance stability and binding.
Researchers say the greatest promise lies ahead: AI-designed antibodies could eventually target disease-relevant proteins that have long been inaccessible to traditional methods, cross the blood–brain barrier, or perform multiple binding tasks in a single molecule. With the ability to generate new antibodies “at the push of a button,” as one scientist put it, drug developers are shifting focus toward the most complex and previously unreachable problems in biology.
