Public launches AI trading agents on Kalshi's prediction market

Gabriel Perez had the speeches before anyone else did. The former White House teleprompter operator read the words before the president said them, bought contracts on whether the president would say them, and collected $172,000 before the CFTC caught up. That settlement, finalised last month, is the clearest illustration of what the Commission's Division of Market Oversight put into writing on Tuesday: mention markets have a structural problem that no surveillance protocol fixes cleanly.

The advisory, signed by acting director Duncan Hennes, establishes a presumption that contracts settling on the discrete conduct of a named individual — a word spoken, an appearance made, a photograph taken — are open to manipulation in ways that contracts on election results or economic data are not. The distinction matters more than it sounds. When a contract resolves on GDP growth, no single actor controls the outcome. When it resolves on whether a specific person says a specific phrase, that person, and everyone with access to their schedule, holds an edge that is both asymmetric and invisible to the market.

The CFTC's four-factor test for rebutting the presumption is workable in theory and mostly unworkable in practice. It asks whether the named individual faces legal or professional deterrents, whether the conduct is independently verifiable, whether they can be pressured by third parties, and how good the exchange's own surveillance is. Run those four tests against a podcast host, a mid-level executive, or a social media influencer with no fiduciary duty to anyone and the rebuttal fails on every factor simultaneously.

Where the analysis gets interesting is Source 4: the same week the CFTC warns that automated advantage-taking in mention markets is manipulation-adjacent, Public has announced a tie-up with Kalshi to bring AI trading agents directly onto the platform. These are not tools that screen for compliance. They are tools that execute faster, at scale, on exactly the informational edges the advisory describes. A human trader who knows a podcast host's run sheet has a window of hours. An AI agent with the same input closes the same trade in milliseconds and cycles through dozens of contracts before any surveillance flag moves.

The advisory carries no legal force. Staff advisories represent the views of a division, not the Commission, and Hennes is acting director, not a confirmed chair. Designated contract markets are not required to delist anything. But the document shifts the burden of proof, and the four-factor standard gives the CFTC a credible hook for enforcement action against any exchange that lists a mention market without a documented surveillance case on file.

The deeper tension is not between the CFTC and the exchanges. It is between what AI-augmented trading makes possible and what a manipulation standard written for human actors can catch. The Perez case was detectable because the trades were placed before a scheduled speech and the position size was legible. A distributed AI strategy running across dozens of mention contracts simultaneously may generate no single flag that a compliance officer recognises as a flag.