What counts as "AI-discovered" on Pipeline Watch
Pipeline Watch is a plain-English tracker of medicines that AI helped find or design and that are now being tested in people. There is no official definition, so here are our rules.
A drug is listed only if a company, a peer-reviewed paper, or reputable trade press says AI or machine learning played a real role in finding the drug's target or designing the molecule. There is no official standard, and the FDA does not label drugs this way, so we show our confidence on every row.
Confidence A (AI-discovered / AI-designed): The developer or a peer-reviewed source says the target and/or the molecule was found or designed by an AI or machine-learning platform (generative chemistry, generative antibody design, AI target discovery).
Confidence B (AI-assisted): AI or machine learning was one tool among others. Examples: physics-based simulation combined with machine learning (Schrödinger, Nimbus/zasocitinib, Relay), AI used to pick a drug or indication (Lantern LP-300), or programs where the AI role is described only by the company or an independent tracker. Shown in the table with an "AI-assisted" label (A rows show "AI-discovered").
Not included: Drugs that merely used standard computer modelling, drugs where AI only helped run the trial, and drugs with no public source for an AI role.
Row rules
- Every row needs a public source URL (ClinicalTrials.gov study page where one exists, otherwise the company press release).
- "Last update" is the ClinicalTrials.gov last-update-posted date unless the "last_update_basis" column says otherwise. A blank date means we could not verify it, and the table shows "Not verified".
- Phase and status are copied from the registry or company source at time of research (2026-10-15). Nothing was estimated.
- Failed and discontinued programs are included on purpose, because they are part of the story.
- This is a seed list, not the complete field. Published counts range from about 117 to 175 AI-originated programs depending on the tracker (e.g., a 2026 ASCO abstract counted 117 AI-enabled assets in human trials; trade analyses say 173+). Pipeline Watch lists 34 so far.
Verified 2026-10-15: Every row was re-checked against its ClinicalTrials.gov record (or ANZCTR for ABS-101) plus a company press release or peer-reviewed paper for the AI basis. Programs whose AI role could not be supported by a primary source were removed.