ByteDance unveils an AI-designed drug candidate against IL-17 for autoimmune diseases

ByteDance publicly showed for the first time how its AI goes beyond video recommendations and enters biomedicine. The Anew Labs division presented a drug candidate for autoimmune diseases — and it concerns a target that pharma has considered too complex for conventional small molecules for decades.

What was shown in Boston

On April 16, 2026, at the annual meeting of the American Association of Immunologists in Boston, the Anew Labs team revealed preclinical data on its leading candidate: an oral small molecule created with generative AI to block IL-17. This signaling protein is associated with psoriasis, rheumatoid arthritis, and ankylosing spondylitis. For ByteDance, this is the first public confirmation that its drug discovery direction has moved out of internal research mode and demonstrates a concrete therapeutic project.

What matters most here is not just the company name, but the type of target. IL-17 in this format pertains to protein-protein interactions, long called undruggable — too "flat" and large for classical small molecules. According to the paper abstract, the Anew candidate can act on multiple IL-17A and IL-17F dimers simultaneously, and in preclinical models showed activity comparable to the antibody bimekizumab.

While this is not yet a market drug, but an early candidate, such steps are considered true tests for AI approaches in pharma.

"These are the first oral small molecules with strong activity against

IL-17F and pronounced in vivo activity," the abstract states.

Why it matters

Today, the most successful IL-17 drugs are mostly injectable biological medicines. They work, but require injections, are more complex to manufacture, and are not always convenient for patients. If ByteDance can truly bring an oral molecule to the clinic and then to the market, this will mean not just another AI case study, but an attempt to replace part of expensive injectable regimens with a pill. For autoimmune diseases, this is a particularly sensitive scenario, because therapy is often needed for a long time.

• Target — IL-17A and IL-17F, key drivers of a range of autoimmune diseases
• Candidate format — oral small molecule, not an injectable antibody
• Design was conducted through AI-enabled structure-based generative design and virtual screening
• Preclinical data presented on binding, cellular activity, and in vivo models

That said, there is plenty of reason for caution. The presented results are conference and preclinical data, not clinical evidence of efficacy in humans. In biopharma, between a strong presentation and an actual drug lies years of optimization, toxicology, regulatory stages, and a high probability of failure. So the news is important not as a finished breakthrough, but as a signal: a major technology company is already demonstrating not general promises, but a concrete molecule for concrete biology.

Why ByteDance

Anew Labs looks not like a random side project within a major tech company, but as a separate ByteDance bet on computational biology. The company website lists teams in Shanghai, Singapore, and San Jose, as well as 36 key employees and a scientific advisory board with people from Innovent Biologics, Amgen, and Takeda. The ambition is clear: tackle targets where conventional drug discovery is bogged down and find therapeutic forms that can be made more accessible and convenient for patients.

This story is important also because ByteDance is building not a single project, but a platform. In March 2026, Anew Labs published a preprint of AnewOmni — a generative system trained on more than five million biomolecular complexes. The authors claim that the model can design functional molecules at different scales: from small compounds to peptides and nanobodies.

The Anew website also lists AnewSampling and other tools, meaning the company is building a full stack for searching, evaluating, and optimizing candidates. Against this backdrop, ByteDance is plugging into an already formed race for drug discovery with AI alongside DeepMind, Anthropic, and Insilico Medicine. The difference is that now the conversation is not about pretty demos and publications alone, but about who can first take a model-generated molecule through the entire development cycle.

For tech companies, this is a long and expensive game, but it is precisely this that separates laboratory hype from real pharma business.

What it means

ByteDance shows that the next major AI battle is not only over search, code, or content, but over physical products with measurable biological effect. If Anew Labs can confirm results in further research, the market will have a strong argument that generative models are beginning to create not just hypotheses, but drug candidates for tasks once considered nearly inaccessible.