OncAdios

Companion artifact

AI-Oncology Failure Watch — Locked Seed List

Companion to How I Would Know If I'm Wrong. Locks the denominator universe behind Falsifier 1. Lock date: May 15, 2026. Universe size: 15 unique companies.

The list is fixed at this date and cannot be expanded retrospectively. Companies can exit (acquired, IPO'd, ceased operations); they cannot be added once a failure event becomes visible. The list is published in full so that the falsifier is testable in public, not asserted in private.

Inclusion criteria — all four required

  1. Stage as of 2026-05-12 — pre-IND or in Phase I (or explicitly publicly targeting that transition within the 18-month window).
  2. Financing — Series A or Series B closed between 2024-01-01 and 2026-05-12.
  3. AI load-bearing — AI material to the regulatory narrative in at least one of: target discovery, candidate prioritization, patient selection, dose optimization, endpoint adjudication. Peripheral computational tools do not qualify.
  4. Public observability — has filed an IND, made a Series announcement covered by trade press, or is otherwise publicly identifiable from a non-paywalled source.

The locked universe

Each entry below names the lead program, stage, Series round and date, amount, and AI role, with a public source link. Companies with multiple qualifying programs are de-duplicated to one universe entry per the operational rule that the rate measures unique companies that failed.

  1. Iambic Therapeutics — IAM1363 (HER2 TKI) + IAM-C1 (CDK2/4 inhibitor); Phase I/Ib + pre-IND; Series B ext., 2024-Q2; $50M ext. ($150M total Series B). AI: target discovery, candidate prioritization. Source.
  2. CHARM Therapeutics — CHM-029 (next-gen menin inhibitor for AML); pre-IND, Phase I targeted Q2 2026; Series B, 2025-Q3; $80M. AI: target discovery, candidate prioritization (DragonFold). Source.
  3. Atavistik Bio — ATV-1601 (allosteric AKT1 E17K-mutant inhibitor); Phase I; Series B, 2025-Q4; $160M. AI: target discovery, candidate prioritization (AMPS). Source.
  4. Auron Therapeutics — AUTX-703 (KAT2A/B degrader for r/r AML/MDS); Phase I; Series B, 2025-Q1; $27M. AI: target discovery, patient selection, candidate prioritization (AURIGIN). Source.
  5. Cartography Biosciences — CBI-1214 (LY6G6D T-cell engager for MSS/MSI-L CRC); Phase I; Series B, 2025-Q4; $67M. AI: target discovery, candidate prioritization (ATLAS, SUMMIT). Source.
  6. Treeline Biosciences — TLN-121 (BCL6 degrader), TLN-372 (pan-KRAS), TLN-254 (EZH2); Phase I (3 trials); Series A ext., 2025-Q3; $200M ext. ($1.1B total). AI: target discovery, candidate prioritization. Source.
  7. Crossbow Therapeutics — CBX-250 (TCR-mimetic T-cell engager for r/r AML/MDS); Phase I; Series B, 2026-Q1; $77M. AI: target discovery, candidate prioritization (T-Bolt + pHLA target ID). Source.
  8. InduPro Therapeutics — Lead proximity-based bispecific oncology candidate; IND filed Q4 2025 → Phase I; Series A, 2024-Q2; $85M. AI: target discovery, candidate prioritization (ProXiMATE). Source.
  9. OnCusp Therapeutics — CUSP06 (CDH6 ADC for platinum-resistant ovarian + solid tumors); Phase I (Phase 1a data ASCO 2025); Series A, 2024-Q1; $100M. AI: target discovery, candidate prioritization (Multitude). Source.
  10. Kivu Bioscience — KIVU-107 (PTK7 ADC) + KIVU-206 (second ADC); Phase I + pre-IND; Series A, 2024-Q4; $92M. AI: candidate prioritization (Synaffix + AI stability modeling). Source.
  11. Sidewinder Therapeutics — SWT012 (bispecific ADC for EGFR co-complex in CRC, SCC); pre-IND (IND end-2026); Series B, 2026-Q2; $137M. AI: target discovery, candidate prioritization (bispecific co-complex algorithm). Source.
  12. Callio Therapeutics — HER2 dual-payload ADC; pre-IND (Phase I 2025/2026); Series A, 2025-Q1; $187M. AI: candidate prioritization (Hummingbird Bio AI). Source.
  13. Cytospire Therapeutics — CYT X300 (pan-γδ T-cell engager for EGFR+ tumors); pre-IND (IND-enabling); Series A, 2026-Q2; $83M. AI: target discovery, candidate prioritization (AI multispecific). Source.
  14. Syneron Bio — Macrocyclic peptide oncology candidate (Synova platform); pre-IND; Series B, 2026-Q1; $150M. AI: target discovery, candidate prioritization (Synova + HTS). Source.
  15. Link Cell Therapies — LNK001 (logic-gated CAR-T for RCC); pre-IND (IND + Phase I 2026); Series A, 2025-Q4; $60M. AI: candidate prioritization, patient selection (logic-gating co-expression). Source.

The falsification threshold

The companion piece commits: "If, eighteen months from publication, fewer than one-third of the companies in that universe have hit one of the three failure patterns... then the rate is overstated."

For the locked universe of 15 companies, the falsification threshold is fewer than 5 failure events, each uniquely attributed to a company in the universe. The threshold is expressed as a fraction rather than an absolute count because the universe may shrink (companies exit). The fraction is the testable rule.

Adjudications made at locking

Five categories of judgment call were made on 2026-05-15. Each is recorded here for the public audit trail.

Excluded — Series stage outside A/B window

  • Pathos AI (Series D), Insilico Medicine (Series E), Dewpoint Therapeutics (Series D), Generate:Biomedicines (Series C, IPO'd March 2026), Genesis Therapeutics (Series B was 2023, outside window).

Excluded — platform-shaped, not biotech-shaped

  • Terray Therapeutics — internal pipeline primarily immunology; oncology exposure via BMS/Calico partnerships.
  • Zephyr AI — patient-stratification platform selling into pharma, not program-owning biotech.

Excluded — AI-load-bearing status unverifiable from public sources

  • Stipple Bio — Pointillist Platform's epitope-identification workflow has no public AI language; reinstatement requires verifiable source.

Excluded — series classification unclear

  • Earendil Labs ($787M March 2026, no series letter assigned) — universe definition requires explicit Series A or B.

De-duplicated — multiple programs, one company

  • Per the operational rule ("rate measures unique companies that failed"), multiple programs at the same company collapse to one universe entry. Affected: Iambic (two programs), Kivu (two programs).

Maintenance rules (after lock date)

The list is fixed at 15 companies. Future quarterly compilations may:

  • Mark exits — a company that is acquired, IPO'd into a non-Series-stage category, or ceases operations is marked "exited" with a date and reason. Its pre-exit period still counts toward the universe denominator for the windows it was present.
  • Annotate — new disclosures, program advances, or stage changes are added as annotations.
  • Refuse retrospective edits — a company that would have qualified at lock date but was missed cannot be added later. A company that no longer meets the criteria after lock date does not exit; it remains in the universe at its locked status.

Open questions for the post-mortem

Tracked at the operational register and revisited at the 12 and 24-month marks:

  • If the universe is too small for statistical signal, the 24-month review will say so plainly. The 18-month window is the falsifier; the 24-month window is the longer-horizon test.
  • Companies that pivot mid-window (e.g., biotech-shaped → platform-shaped) are tracked separately in compilation notes.
  • The locked universe does NOT cover EMA, PMDA, or NMPA failures. The agency-specific scope is FDA. Multi-agency broadening, if warranted, is a v2 decision.

← Back to How I Would Know If I'm Wrong · Writing index