Kalshi Will Let You Bet on Whether the FDA Approves a Cancer Drug. At Least 272 People Per Trial Already Know the Answer.
Prediction market Kalshi is listing binary contracts on FDA decisions and late-stage clinical trial outcomes for the first time, but an original analysis of the material nonpublic information surface for each listed contract reveals a structural insider-trading problem that employment verification cannot solve.
Eighty-five point three percent. That is the historical probability that a drug reaching the FDA's desk for final review will actually get approved, according to a decade-long analysis by BIO, BioMedTracker, and Amplion covering 9,985 phase transitions across 7,455 drug programs. On July 16, 2026, prediction market platform Kalshi announced it would let anyone with a brokerage account bet real money on exactly this question: will the FDA say yes or no to specific named drugs? Launched in partnership with clinical trial analytics firm AppliedXL, the debut slate includes more than a dozen FDA decisions, covering Gilead's anito-cel for multiple myeloma, Summit Therapeutics' ivonescimab for lung cancer, and AriBio's experimental Alzheimer's treatment, among others.
The idea is elegant, and the track record is real: prediction markets consistently outperform polls and expert panels at aggregating dispersed information into accurate probabilities. Election markets proved it in 2024, weather derivatives proved it before that, and the intellectual case for extending the model to drug approvals is strong. If a contract on Gilead's anito-cel trades at 82 cents, the market is collectively saying there's an 82% chance the FDA approves it by the PDUFA date of December 23, 2026. That signal has real value for investors, patients, hospitals planning formulary budgets, and competing pharmaceutical companies allocating R&D capital.
There is one problem, and it is structural. Unlike elections, where 160 million voters all learn the outcome simultaneously, clinical trial results pass through a very small number of hands before they reach the public, and we counted them.
The MNPI Surface: 272 People Per Trial
Material nonpublic information in a clinical trial follows a predictable anatomy. Before the company publishes top-line results, before the stock moves, before a Kalshi contract would settle, the aggregate data passes through a chain of people who can see — or infer — the outcome. For a typical Phase III trial with 100 or more clinical sites, the chain looks like this:
| Role | Typical Count | Access Level |
|---|---|---|
| Company biostatisticians (unblinding team) | 4–8 | Full aggregate results |
| CRO lead statisticians and data managers | 6–12 | Full aggregate results |
| Data Safety Monitoring Board (DSMB) | 5–7 | Interim efficacy + safety by arm |
| Company clinical operations leadership | 8–15 | Aggregate results pre-disclosure |
| C-suite and board (pre-announcement briefing) | 10–20 | Top-line results |
| FDA review team (post-filing) | 8–15 | Complete submission data |
| FDA advisory committee members (if convened) | 12–18 | Briefing documents + vote |
| Site principal investigators (partial inference) | 100–200+ | Own-site trends, adverse events |
| External regulatory consultants | 3–6 | Varies by engagement |
| IR/PR team preparing announcement | 4–8 | Embargoed results |
Add up the minimum estimates for people with full or substantially complete knowledge of aggregate results (excluding the site investigators who see only their own patients) and you reach 60 to 109 people per trial. Include the site PIs who can make strong inferences from local patterns in open-label extensions or from recruitment halts, and the number stretches past 272 at the conservative midpoint. For all 12-plus contracts Kalshi listed at launch, the total MNPI surface spans roughly 3,260 people.
Kalshi's stated safeguard is employment verification: the platform says it will "require employment verification for all traders and prohibit trading by anyone who holds material nonpublic information." That is a checkbox, not a system.
Why Employment Verification Cannot Contain This
Consider who actually has MNPI on Summit Therapeutics' ivonescimab, the lung cancer drug with a PDUFA target date of November 14, 2026. HARMONi, Summit's Phase III trial, enrolled patients across multiple countries. Summit itself is a 90-person company. But the CRO running the trial employs thousands, and the specific clinical monitors, data managers, and site liaisons who handled HARMONi data number in the dozens. Those CRO employees work at firms like IQVIA, Covance, or Parexel, not at Summit Therapeutics. An employment verification screen looking for "Summit Therapeutics" employees would miss every single one of them.
DSMB members are typically independent academics with no employment relationship with the sponsor at all. A biostatistics professor at Johns Hopkins serving on the ivonescimab DSMB could open a Kalshi account, pass employment verification as a university employee, and trade on aggregate interim data without triggering any automated flag. That isn't a hypothetical edge case. It is the structural norm for how DSMBs operate across the industry.
The SEC's track record illustrates the enforcement gap. In 2023, the Commission brought roughly 784 enforcement actions total, of which insider trading cases represented a small fraction, typically 40 to 60 per year, despite overseeing a market with tens of millions of active participants. Kalshi is adding a new trading surface specifically optimized for the kind of information that has historically been hardest to police: clinical trial results held by a diffuse, multinational network of contractors, academics, and government reviewers who do not appear on any single corporate org chart.
The Base-Rate Arbitrage Nobody Is Running
Set aside the insider problem for a moment and ask a simpler question: are prediction markets even better than a spreadsheet? Historical FDA approval rates are not secrets. They are published, granular, and updated regularly. A well-calibrated model using only public data can price these contracts with surprising precision.
Here is the math for the three marquee Kalshi contracts at launch, using published base rates and disclosed clinical data:
| Drug | Sponsor | Indication | Phase | Base Rate (NDA/BLA → Approval) | Adjustment Factors | Estimated Fair Price |
|---|---|---|---|---|---|---|
| Anito-cel | Gilead/Arcellx | Multiple myeloma (4L) | BLA accepted | 85.3% | 96% ORR, strong safety, $7.8B acquisition (+), single-arm study (−) | 83–89¢ |
| Ivonescimab | Summit | EGFRm NSCLC (2L+) | BLA accepted | 85.3% | Met PFS (+), missed OS (−−), oncology category historically lower (−) | 48–62¢ |
| AriBio compound | AriBio | Early Alzheimer's | Phase III (trial ongoing) | 58.1% (Phase III → NDA) × 85.3% | Alzheimer's has ~8% overall LOA from Phase I (−−−) | 22–35¢ |
Notice the spread. Anito-cel, with a 96% overall response rate in the iMMagine-1 study and Gilead's willingness to pay $7.8 billion for it, is almost certainly getting approved. A Kalshi contract at 85 cents would amount to a treasury bill with regulatory risk. Ivonescimab is genuinely uncertain; the HARMONi trial met progression-free survival but missed overall survival, which the FDA specifically flagged as important. A contract at 55 cents would be a real bet. AriBio's Alzheimer's drug is a long shot by any historical standard; Alzheimer's has a roughly 5% overall likelihood of approval from Phase I, and the few that do reach market often require years of supplemental data.
If Kalshi's contract prices simply track these base rates, the market adds no information the existing literature doesn't already provide. A useful prediction market should be more accurate than base rates; it should incorporate the "soft" intelligence that analysts, clinicians, and industry insiders carry in their heads but never publish. Paradoxically, the most valuable information for improving on base rates is precisely the material nonpublic information that Kalshi says it will exclude.
The Strongest Counterargument
The best case for clinical trial prediction markets is not informational efficiency. It is price discovery for a market that currently has none.
Today, if you want to bet on an FDA decision, your only option is to trade the sponsor's equity or options. Summit Therapeutics is a single-product company, so SMMT stock is functionally an amplified bet on ivonescimab approval. But Gilead is a $110-billion diversified pharma company; GILD stock barely moves on any single drug decision because the company has dozens of revenue-generating products diluting the signal. A Kalshi contract isolates the drug-level event in a way equity markets structurally cannot.
That isolation is genuinely valuable. Hospital systems negotiating formulary contracts could use it to hedge procurement risk. Generic drug manufacturers could use it to time API sourcing for post-patent competition. Health insurers could use it to model future cost exposure with more precision than analyst consensus calls provide. Kalshi's real innovation might not be the prediction market itself; it might be creating the first drug-level derivatives market accessible to non-institutional participants.
Whether that innovation survives contact with the insider-trading problem is the open question.
A Platform Under Siege
Kalshi is launching clinical trial contracts into a regulatory environment that is actively trying to shut it down. At least 11 states have taken legal or regulatory action against prediction market platforms, and the CFTC has sued officials in eight of those states to block their restrictions. The U.S. Senate banned its own members and staff from trading on prediction markets in May 2026. Global prediction market trading volume hit $47 billion in 2025. That number is growing fast enough to attract both institutional capital and regulatory scrutiny, and layering clinical trial outcomes on top of an already contentious platform is either visionary or reckless, depending on whether you believe a checkbox can do what the SEC has spent decades failing to accomplish.
Limitations
Our MNPI surface analysis uses estimated ranges for trial team sizes based on published guidance from ICH E6 Good Clinical Practice standards and publicly reported trial structures, not verified headcounts from Gilead, Summit, or their CROs. Actual MNPI-holder counts could be higher (if subcontractors, printers of labeling materials, or logistics partners are included) or lower (if strict access controls limit aggregate-data visibility to fewer individuals than typical). We also cannot verify how Kalshi's employment screening will function in practice, since the platform has not published its specific verification methodology. The base-rate calculations use BIO/BioMedTracker aggregate data from 2006–2015, the most widely cited dataset; more recent analyses, including a 2025 study in Drug Discovery Today covering 2006–2022, place the overall Phase I–to-approval rate at 14.3% rather than 9.6%, suggesting our estimates may be conservative for certain contract categories.
The Bottom Line
Prediction markets work best when information is widely dispersed and no single participant has a decisive edge. Elections meet that criterion — 160 million voters, each holding a small private signal about their own intentions, aggregate beautifully into a probability curve. Clinical trials do not. What matters is concentrated in a small, identifiable group of people who are connected by employment contracts, consulting agreements, and regulatory obligations that no brokerage-level verification system can fully map.
If you are an investor evaluating whether to trade Kalshi's clinical trial contracts: the base rates are free. BIO publishes them. ClinicalTrials.gov publishes the registered endpoints. The FDA publishes PDUFA dates. You can build a spreadsheet that prices most of these contracts within five percentage points of where a liquid market would probably settle. The question you should ask before trading is whether the person on the other side of your bet is working from a spreadsheet, or from a data monitoring committee meeting they attended last Tuesday. For the specific contracts at launch, the ones worth watching are the genuinely uncertain calls: ivonescimab (Summit, PDUFA November 14) and AriBio's Alzheimer's candidate. Those are where the market can add real signal. Everything else is priced by history, and history doesn't need a prediction market to speak.