Field Dispatch 07 prediction markets go mainstream

The model calls it.
The chain proves it.

Meta is reportedly building Arena - a play-money prediction market where AI writes the questions, recommends the bets, and has the final say on what happened. Flashcast already runs the same idea with one difference that changes everything: the outcome is not judged. It is proven on-chain.

The modelAI adjudication
Question loading…
reading sources about the event
weighing what probably happened
issuing a judgment call
the model decides
whether it happened
The chainFermah Kernel
Question loading…
event fires on-chain
condition checked against the chain
proof attested on-chain
proven, not judged
no dispute window
Illustration of the two settlement models - not live Arena or Flashcast data.
$130B+sector volume already in 2026, vs $50B in all of 2025
70+companies with prediction-market projects (Event Horizon)
100Mmonthly users Meta reportedly targets for Arena
30+lawsuits pending over the legality of the category
The story

The biggest company in social just validated the category.

Per reporting by The New York Times and NPR, Mark Zuckerberg has made Arena a priority project: play money on real-world outcomes, Llama generating questions from trending topics, personalized market recommendations, a target of 100 million monthly users aged 18 to 34, and talks explored with Polymarket and Kalshi. Later reporting describes plans to fold it into Facebook, Instagram and WhatsApp - betting as conversation.

Buried in the internal documents is the detail that matters most: the same AI that writes and recommends the markets also resolves them, in near real-time. The model has the final say over whether something did or did not happen.

Arena is unannounced. Everything above is sourced from NYT and NPR reporting on internal documents; Meta has declined to comment, and the app may change or never ship.

2020
Meta tries this once
Forecast launches - a crowdsourced prediction app on points. It is shut down two years later; keeping markets accurate by hand proved too expensive.
2024
The election makes the category
Prediction markets break into the mainstream as traders use event contracts to read the US election in real time.
2025
The volume arrives
Kalshi and Polymarket clear a combined $50B in annual trading volume. Robinhood, DraftKings, FanDuel and dozens more pile in.
May 2026
The trust problem surfaces
The CFTC and DOJ bring the first insider-trading case involving a private company employee betting on prediction markets. By July, Goldman Sachs, Morgan Stanley and Bank of America are writing internal policies for employee trading on these platforms.
June 2026
Meta comes back for the whole category
Arena surfaces: AI-written, AI-recommended, AI-resolved markets on play money, aimed at 100M monthly users. Sector volume has already passed $130B this year; Bernstein projects up to $1T by 2030.
Meanwhile
Flashcast is already live
A question becomes a market in about three seconds, and Fermah Kernel settles it on-chain the moment the event occurs. No committee, no dispute window - and no model deciding what is true.
2025 · Kalshi + Polymarket volume$50B
2026 so far · already past$130B
2030 · Bernstein projectionup to $1T
Projection, not a fact - shown hatched for a reason.
Side by side

Same idea. Different judge.

Both run on points, not cash. Both turn questions into markets. The split is at the last step - who decides what actually happened.

Dimension
Arena as reported by NYT / NPR
Flashcast live on testnet, by Fermah
Stakes
Play money, a daily virtual allotment
Testnet points
Who writes the questions
Llama, generated from trending topics
Anyone - a question becomes a live market in about 3 seconds
Who picks what you see
Personalized AI recommendations
The board - markets ranked by volume staked
Who settles the outcome
The model, in near real-time - AI has the final say
Fermah Kernel - the outcome is proven on-chain, not judged
Dispute window
Not described in reporting
None by construction - a proof is not arguable
Status
Internal testing - may change or never ship
Live in private beta today

Arena column reflects reporting by The New York Times and NPR on internal Meta documents (June 2026) plus later reporting on ecosystem integration. Meta has not announced the product.

The wider board

Every platform answers the same question differently.

Strip away the branding and each prediction market is defined by one design choice: who is the judge. Here is how the four main answers look side by side.

Platform
Stakes
Who settles the outcome
Dispute path
Kalshi
Real money, CFTC-regulated exchange
People, applying the exchange rulebook market by market
Structured review inside the exchange
Polymarket
Real money, crypto
An optimistic oracle: a proposed answer stands unless someone challenges it
Challenges escalate to a token-holder vote, adding a dispute window
Arena reported
Play money, daily allotment
The model - AI has the final say, in near real-time
Not described in reporting
Flashcast
Testnet points
Fermah Kernel - the outcome is proven on-chain, not judged
None by construction - a proof is not arguable

Committees, oracles with dispute windows, a model with final say, or a cryptographic proof. More than 70 operators have entered the category, and every one of them lands somewhere on this line.

The regulatory file

The trust problem is now a legal problem.

The judge question stopped being philosophical this spring. A running file:

May 2026
First private-sector insider case. The CFTC and DOJ charge a Google employee with trading Polymarket contracts on material non-public information - the first such case involving a private company's employee.
June 2026
Congress reacts to Arena. Sen. Richard Blumenthal criticizes Meta's reported plans, pointing to the Prediction Markets Security and Integrity Act, which would restrict officials from profiting on inside information.
July 2026
Banks write the rules themselves. Goldman Sachs issues explicit directives on employee prediction-market trading; Morgan Stanley, JPMorgan and Bank of America confirm policies or updates in progress.
Ongoing
Legal limbo. More than 30 lawsuits are pending over the legality of prediction markets, and the CFTC has what legal scholars call a blank canvas on how it will pursue insider-trading cases in the category.

A market where the settlement itself is a verifiable proof has a shorter answer to every row above: there is no judge to lean on.

Why the judge matters

A judged outcome can be argued with. A proven one can't.

Trust
Models are persuadable
An AI verdict is a judgment call: sources can be wrong, edge cases ambiguous, outputs contested. A cryptographic proof of an on-chain event leaves nothing to argue about.
Incentives
The house shouldn't referee
When one company writes the questions, recommends the bets, and rules on outcomes, it is player and referee at once. Settlement by proof takes the referee out of the building.
The moment
Regulators are circling
Insider-trading cases and bank policies show the sector's trust problem is real. Markets that settle by verifiable proof answer the question before it is asked.
Meanwhile, live right now

Flashcast isn't a plan. It's a board.

The markets below are pulled from Flashcast's public data - crypto, social, sports, whatever people are asking about right now.

Top markets by volume Snapshot
Source: flashcast.social public data · snapshot Create a market ↗
Proof Desk / 07

Meta is teaching a billion people to ask questions. Someone still has to prove the answers.

The category is arriving either way. The difference worth betting on is whether truth gets decided by a model with final say - or settled by a proof no one can argue with.

Sources

  1. NPR - Meta plans to release AI-powered prediction market app, internal documents (Jun 24, 2026). npr.org
  2. The New York Times - first report on Arena (Jun 2026), as recapped by TechRepublic. techrepublic.com
  3. Built In - What is Meta's Arena? Integration and "betting as conversation" (Jul 8, 2026). builtin.com
  4. CNBC - Prediction markets spark insider trading concerns; Goldman, Morgan Stanley, BofA respond (Jul 9, 2026). cnbc.com
  5. Yahoo Finance - Arena report; sector volumes $50B (2025) to $130B+ (2026), Bernstein $1T projection. finance.yahoo.com
  6. BigGo Finance - 100M MAU target, 18-34 audience, partnership talks with Polymarket and Kalshi. finance.biggo.com
  7. PlayUSA - lawmaker reaction, Prediction Markets Security and Integrity Act (Jun 29, 2026). playusa.com

Arena is unannounced; all Arena details are drawn from the reporting above on internal documents, and Meta has declined to comment. Flashcast and Fermah Kernel details reflect fermah.xyz and flashcast.social as of July 10, 2026. Settlement descriptions for Kalshi and Polymarket summarize their publicly documented resolution processes.