Will AI Find a Cancer Cure Before Your Doctor Does?
Will an AI-discovered drug receive FDA approval for a previously untreatable condition by end of 2028?
If you or someone you love faces an untreatable condition, this is the most hopeful and most uncertain frontier.
Your Prediction
Where do you think this lands?
Join others who've weighed in
Scenarios
Current value: ~15 AI-designed drugs in Phase III trials (Jan 2026). Most advanced: zasocitinib (TAK-279). No FDA approvals yet.
S-curve position: Very early — proving concept phase, years from mainstream adoption
0 approvals (90% trial failure rate applies equally, 'previously untreatable' qualifier is very narrow)
0-1 approvals (most likely for less restrictive 'any condition' vs 'previously untreatable')
2+ FDA approvals (fast-track designations accelerate timeline, especially oncology)
How We'll Know
- What we measure
- Whether an AI-discovered drug (where AI was primary in target identification or molecular design) receives FDA approval for a condition with no prior approved treatment
- Confirmed if
- An AI-originated drug receives FDA approval for a previously untreatable condition by end 2028
- Refuted if
- No AI-originated drug receives FDA approval for any condition by end 2028
- Data sources
- FDA new drug application filings
- ClinicalTrials.gov
- Nature Medicine / Nature Biotechnology
- Insilico Medicine, Recursion, Isomorphic Labs filings
Evidence Trail
Evidence For
- Mar 7, 2026
IsoDDE doubles AlphaFold 3 accuracy. Chai-2 achieves 16-20% antibody hit rate (100x improvement). ~15 AI drugs in Phase III. Phase III success rate ~50-60%. Boltz-2 1000x faster binding simulations. Pipeline math supports multiple positive readouts by 2028.→ Probability: 20%
- Mar 7, 2026
AI drug pipeline growing 50% annually. Chai Discovery (solo founder Joshua Meier) producing pharma-grade results. The small-team AI pattern applies to drug discovery too — fewer researchers needed for more candidates. Inference cost collapse means more computational experiments per dollar. But biology remains the bottleneck.→ Probability: 25%
- Mar 9, 2026
AlphaGenome (Nature, Jan 2026): first AI analyzing 1M base pair DNA sequences at single-letter resolution, predicting 5,930 genome tracks. Called 'the AlphaFold of genomics.' ~3,000 scientists in 160 countries, ~1M API calls/day. Microsoft MAI-DxO: 85.5% accuracy on complex medical cases vs 20% for physicians. Insilico ISM001-055: positive Phase IIa for IPF. Generate:Biomedicines GB-0895: Phase 3 with ~1,600 patients. FDA authorized 1,356 AI medical devices total. 2026 described as 'the year of Phase III AI drug readouts.' AI compressing early discovery by 30-40%.→ Probability: 30%
- Jun 7, 2026
Google Co-Scientist (a Gemini-based multi-agent hypothesis system) was published in Nature May 19-21 with six independent wet-lab validations across ALS, liver fibrosis, cellular aging and more; Stanford used it to repurpose FDA-approved Vorinostat for liver fibrosis with a 91% reduction in TGFB-induced chromatin changes in organoids, and its ERA tool beat the CDC's own COVID forecasting ensemble. This validates AI hypothesis generation as peer-reviewed science — but repurposing an approved drug is a softer bar than the de-novo, previously-untreatable FDA approval this question measures.→ Probability: 27%
Evidence Against
- Mar 7, 2026
90% of drugs fail clinical trials regardless of discovery method. 'Previously untreatable' is an extremely high bar — most AI drugs target known conditions. FDA approval timeline extends well beyond Phase III completion. AI improves target identification but doesn't bypass biological complexity.
How Our View Evolved
- Jun 7, 202625%↑27%
+0.02 to 0.27. Google Co-Scientist published in Nature (May 19-21) with six independent wet-lab validations — Stanford repurposed Vorinostat for liver fibrosis (91% reduction in TGFB chromatin changes) and ERA beat the CDC's COVID forecasting ensemble. Strongest evidence yet that AI generates externally validated hypotheses, though this is repurposing of approved drugs, not the de-novo previously-untreatable approval the resolution requires.
- Mar 15, 202630%↓25%
Audit recalibration: mapped predictions (P-004, P-008) address broad AI discovery/medicine, not FDA approval for untreatable conditions. 90% trial failure rate + 'previously untreatable' bar make 2028 very tight. Slight downward correction.
- Mar 9, 202625%↑30%
AlphaGenome breakthrough (Nature, Jan 2026). Phase III AI drug readouts in 2026. FDA authorized 1,356 AI medical devices. Pipeline accelerating faster than expected.
- Mar 8, 2026Initial assessment: 25%
Baseline — initial published assessment
What Experts Say
Sam Altman
CEO, OpenAI
“AI systems will be able to discover genuinely novel scientific insights in 2026”
Demis Hassabis
CEO, Google DeepMind; Nobel Laureate
“Medicine won't look like it does today in 10-15 years; AI will enable personalized treatments and cure major diseases”