Ataraxis AI says causal model predicts chemo benefit across solid tumors

Ataraxis AI said its causal “world model” Ataraxis Tau predicted which patients benefit from adjuvant chemotherapy, reporting three breast-cancer validation studies covering 18,426 patients — including analyses of Phase 3 trials TAILORx and UNIRAD — and saying the same model transferred zero-shot to other solid tumors in 6,692 patients across 16 cancer types.

A cancer-AI company is claiming it can say whether chemotherapy helps this patient, not only how risky the tumor looks. That claim is a same-day company wire with Phase 3 trial analyses — company-said, not independently confirmed by Reuters or AP.

Ataraxis AI, a cancer-AI startup, said via Business Wire on 14 Sep 2026 at 11:00 ET that it had validation results for Ataraxis Tau, a causal AI “world model.” Causal AI, as the company uses it here, means software that tries to simulate what happens to one patient with a treatment and without it. That company release is the filing event. These are company claims. This desk did not rerun the models.

The first application, the company said, is hormone-receptor-positive, HER2-negative early breast cancer — a common early-stage subtype written HR+/HER2−. The job is guessing who benefits from adjuvant chemotherapy. Adjuvant chemotherapy, in plain English, is extra chemo given after the main treatment, usually surgery, to try to kill leftover cancer cells. File that as Ataraxis’s first-use claim.

The company said three breast-cancer validation studies covered 18,426 patients, including analyses of two Phase 3 trials — large late-stage clinical trials — named TAILORx and UNIRAD, plus a real-world study. File the 18,426 total and the trial names as company-stated.

In the TAILORx analysis, Ataraxis said the model was predictive of adjuvant chemotherapy benefit, with an interaction p-value of 0.016 — a company-stated statistics line meaning the company reads a treatment-by-model interaction as unlikely by chance. File p=0.016 as the company’s figure. This desk did not recompute it.

In the UNIRAD analysis, the company said the model was predictive of benefit from everolimus — an mTOR inhibitor, a targeted drug — in an identified subgroup, with p=0.01, company-stated. File that p-value as Ataraxis’s. This is not a new UNIRAD result from the original trial team.

The real-world breast study, the company said, covered 11,135 patients from 17 cohorts in 10 countries, and the model was predictive of chemotherapy benefit (p<0.001, company-stated). File the 11,135 / 17 / 10 and the p-value as Ataraxis’s.

The same model, Ataraxis said, transferred zero-shot to other solid tumors. Zero-shot, in plain English, means the company applied the breast-trained model to cancer types it was not trained on, without extra fine-tuning. The company reported 6,692 patients across 16 other cancer types; prognostic — guessing how the disease may go — in 11; and chemotherapy-benefit transfer in 5 of 7 cancers treated with platinum- or taxane-based regimens, drugs also used in breast chemo. File those as company claims.

Jan Witowski, MD, PhD, co-founder and CEO, said today’s results mark a major milestone and that Ataraxis Tau can predict recurrence and treatment benefit across cancer types. File the quote as his, via the company release. This desk is not saying the model is ready for every cancer patient.

Ataraxis Breast, the company said, is a suite of AI tests that combine routine digital pathology — scanned microscope slides — with clinical data. Tests are offered as laboratory-developed tests, or LDTs, through a CLIA-certified and CAP-accredited lab. CLIA and CAP are U.S. lab-quality certifications; they are not FDA clearance. The company’s research page names Ataraxis Breast CTX as the first Tau-based test for individualized chemotherapy-benefit prediction in HR+/HER2− breast cancer. File Breast / Breast CTX as company product names. This desk is not saying the FDA cleared these tests.

PRIMARY here: Ataraxis AI’s 14 Sep 2026 Business Wire release at 11:00 ET — company source. Not independently confirmed by Reuters or AP. The Financial Content page is a same-day Business Wire mirror. The Ataraxis research-causality page is background only. NOT claimed: independent replication, survival guarantees, that this desk reviewed the underlying papers, FDA clearance, a stock tip, or investment advice. Distinct from the already-filed oracle-health-clinical-ai-agent-nurses, tandem-health-100m-series-b, beta-bionics-mint-fda, and intuitive-davinci-sp-ce-europe.