preferred on
Points of Focus
- Kalshi filed contracts on whether OpenAI and other AI labs halt model development or deployment over safety risks.
- Qualifying risks include catastrophic harm, cyber threats, loss of control and breached safety thresholds.
- The filing creates a reusable contract class covering developers, models and model families.
Kalshi has filed a new class of prediction markets tied to whether OpenAI and other artificial intelligence developers halt model development or deployment because of safety risks.
Commodity Futures Trading Commission records show two KalshiEX self-certifications dated Sept. 14. One covers OpenAI specifically. The second establishes a broader contract class that can be applied to a named AI developer or group of developers, a model or model family, a development or deployment activity, and a specified time period.
The broader filing names OpenAI, Anthropic, Google DeepMind, Meta, xAI, Microsoft, Amazon, Mistral, DeepSeek and other developers among its source agencies. Kalshi said the contracts would be listed on a custom basis.
The filings expand prediction markets further into AI. Crypto.com said earlier this month that it would launch more than 20 AI contracts tracking consumer and enterprise adoption.
Kalshi sets detailed rules for an AI safety halt
Kalshi defines development as pre-training, training, post-training, fine-tuning, scaling and capability-advancing research or engineering. Deployment includes release through an API, consumer interface, enterprise service or publication of model weights.
A qualifying halt can be temporary or indefinite, provided the developer announces an actual stop or suspension. A release delay can also qualify when deployment is suspended until a stated safety condition, evaluation or safeguard is completed.
Qualifying risks include severe or catastrophic harm, chemical, biological, radiological or nuclear weapons capabilities, offensive cyber capabilities, loss of human control, breached safety thresholds and failed safety evaluations or controls. Commercial considerations, compute shortages, supply-chain constraints, personnel changes, litigation and ordinary scheduling changes are excluded when they are the sole reason for a halt.
The contracts use binary settlement. A qualifying announcement produces a $1 payout on the Yes side. Kalshi’s source hierarchy begins with official company websites, safety publications, regulatory filings and public statements, followed by designated news organizations.
Market-integrity concerns have grown alongside more specialized event contracts. A recent cluster of Polymarket accounts won 41 of 42 KPMG-linked earnings markets, while ESMA has warned of insider trading and manipulation risks.
Class filing follows CFTC warning on broad templates
The OpenAI product was submitted under CFTC Regulation 40.2(a). The wider AI-lab series uses the class-certification process under Regulation 40.2(d).
Self-certification allows a designated contract market to list a product without prior CFTC approval after certifying compliance with the Commodity Exchange Act and Commission rules. CFTC listing procedures describe that process explicitly.
The distinction has gained attention as exchanges file contracts across new categories and states challenge the CFTC’s federal oversight model, producing competing state and federal claims over event contracts.
In July, CFTC staff warned exchanges about broad template-style event-contract certifications that combine permutations with different settlement sources or methodologies. CFTC staff said those filings can make regulatory review more difficult.
Kalshi cites that advisory directly. Its filing says every contract in the AI-safety class will use the same source hierarchy, payout formula and settlement methodology, while the developer, model, activity and time period serve as variables.
Share
Most read articles
