Regulators Turn to AI to Police Prediction Markets

The Commodity Futures Trading Commission plans to deploy AI surveillance tools to monitor prediction markets like Kalshi and Polymarket, Ars Technica reports. The pitch is that these event-contract platforms have grown faster than the agency's traditional oversight capacity, and pattern-matching software can flag manipulation on elections, sports, and policy outcomes that humans would miss.

It is a reasonable response to a real problem. Polymarket volumes spiked into the billions during the last U.S. election cycle, and Kalshi has been winning court battles to list contracts on political outcomes. Those markets attract exactly the kind of informed-trader edge that surveillance regimes exist to police: a congressional staffer, a campaign pollster, or a sports insider can move prices well before public information catches up.

The harder question is whether AI surveillance can keep up with AI-assisted trading on the other side. Generative models are already being used to scrape sentiment, parse filings, and coordinate small accounts. A regulator's anomaly detector trained on yesterday's manipulation patterns will struggle against tomorrow's automated wash-trading rings.

Key points

• CFTC plans to use AI to detect insider trading and manipulation on prediction markets
• Kalshi and Polymarket have grown rapidly on elections, sports, and policy contracts
• Traditional surveillance tools were built for futures and commodities, not event contracts
• AI-versus-AI dynamics will define enforcement effectiveness

The surveillance stack the agency is implicitly building looks like this:

Prediction Market Venues
   (Kalshi, Polymarket)
          │
          ↓
  Trade + Order Book Feed     ← timestamps, account IDs
          │
          ↓
   AI Anomaly Detection       ← clustering, sequence models
          │
     ┌────┴────┐
     ↓         ↓
  Routine    Flagged Pattern
  Activity         │
                   ↓
           Human Investigator
                   │
              ┌────┴────┐
              ↓         ↓
          Dismissed   Enforcement
                       (fines, bans,
                        referrals to DOJ)

There is a structural tension here that the CFTC has not addressed publicly. Prediction markets are valuable precisely because informed traders bring private information into prices. That is the entire epistemic case for them. The line between informed trading, which markets want, and insider trading, which regulators ban, is murkier on a Polymarket contract about a Supreme Court ruling than it is on pork bellies. A clerk who knows the decision and a court watcher who guessed correctly will look similar in the order book.

Expect early enforcement to target the easy cases: coordinated accounts, obvious timing around non-public announcements, employees of polling firms trading their own data. The harder cases, where the AI flags a statistical anomaly but cannot articulate what rule was broken, will test whether the agency's new tooling produces evidence that holds up in court or just generates a pile of unactionable alerts.

If the CFTC succeeds, it will set a template other agencies copy. If it fails publicly, prediction markets will gain another argument for being treated as something other than derivatives.