Blanket, Kalshi, and the Quiet Limits of AI Risk Advisory
Tracing the static in the protocol’s genesis block is a habit I cannot break. When Kalshi announced on August 7 that it had launched Blanket, a third-party AI-powered risk analysis tool, the static was not in the code. It was in the framing. Blanket is not a blockchain protocol, not a token, and not even an official Kalshi product. It is an independent project by Lauris Zminsky, a fintech entrepreneur, designed to help small businesses map weather, energy, tariff, and election risks to event contracts on Kalshi’s CFTC-regulated marketplace. The product does not execute trades or touch funds. It recommends. That is both its strength and its trap.
To understand why Blanket matters, you first have to strip away the AI gloss and stare at the regulatory architecture. Kalshi is a designated contract market registered with the Commodity Futures Trading Commission. It is centralized, licensed, and stubbornly real-world. It holds user funds, settles event contracts, and publishes clear rules. There is no native token, no yield farming, and no governance community. In crypto terms, it is the boring part of a market that most DeFi natives pretend does not exist. Blanket sits one layer above Kalshi, in the application and tools layer. It reads market data and external risk signals, then presents suggestions in a language a small business owner can understand. The actual transaction, if one happens, occurs on Kalshi.
This division of labor is not accidental. It keeps Blanket out of custody, out of execution, and thus out of the most dangerous regulatory categories. In the events industry, that is what passes for security. Security is a silent promise kept between nodes, and here the nodes are legal entities, not validators. The workflow is simple enough to describe in one breath: a business owner types in an exposure, Blanket parses the language, matches it against an event contract, estimates the implied probability and potential payout, and then remains silent while the user crosses over to Kalshi to act. That boundary matters more than any model accuracy metric. It is the difference between being a tool and being a broker.
Now we enter the part of the analysis that most press releases will not tell you. I have spent enough time inside smart contracts since the 2017 audit cycle to know that the word AI on a product page usually means a rule engine wearing a language model costume. From what has been disclosed, Blanket does not publish latency figures, accuracy benchmarks, or any third-party audit of its recommendation engine. That does not mean it is fraudulent. It means the product has not yet earned the technical trust that infrastructure projects are forced to demonstrate. The innovation here is combinatorial, not structural. Each component inside Blanket is mature: Kalshi provides the marketplace, external data providers supply weather and macro inputs, and a language model synthesizes the output into an advisory interface. The newness is in the intersection, not the invention. That is a legitimate form of innovation, but it carries a fragile assumption: that the underlying event contracts have enough liquidity to make a recommendation useful.
Kalshi’s public contract coverage appears broad enough for initial categories such as weather, energy, tariffs, and elections. But depth is undisclosed. A recommendation to hedge a cold snap with a weather contract only works if there is a counterparty on the other side at a reasonable spread. In thin markets, every AI recommendation becomes a slippage puzzle. The product does not solve that puzzle. It inherits it from the venue. This is why I treat Blanket as an application experiment rather than infrastructure. It sits on top of someone else’s liquidity, someone else’s custody, and someone else’s legal liability. The only thing it owns is the advice. And advice is the hardest asset to defend.
Over the years, I have watched oracle feed latency destroy leveraged positions in DeFi pools, usually in the time between a price update and a liquidation engine’s reaction. Kalshi’s settlement process is a centralized oracle in the purest sense. The CFTC-approved exchange decides what happened, when it happened, and what the payout is. For a small business hedging against a tariff change or an election outcome, that centralized oracle is not a compromise. It is the entire value proposition. The legal certainty replaces the need for trustless consensus. This is why Blanket can never be judged by the same standard as a decentralized prediction market. It is not trying to replace trust with code. It is trying to make regulated trust easier to consume.
The token economist in my head finds the absence of a token refreshing. Or rather, the absence of a token is the most important design decision in the entire project. Blanket has no token, Kalshi has no token. There is no supply schedule, no unlock event, no treasury, no inflationary reward. There is no Ponzi risk because there is no future commitment being sold to present users. Kalshi captures value through transaction fees on event contracts, likely the only sustainable revenue line in the stack. Blanket’s business model remains undisclosed. It might charge a subscription, collect a referral fee, or operate as a loss-leading portfolio piece for its developer. Because there is no token, the standard crypto valuation framework is not just unhelpful; it is misleading. Blanket is closer to a fintech SaaS dashboard than to a Web3 primitive. The only sensible way to evaluate it is to ask whether it can acquire and retain paying business users in a category where most small enterprises have never traded a derivative.
The market context makes that question more interesting than the product itself. Value flows where attention decides to rest, and right now attention is no longer resting on election contracts. The 2024 cycle created a surge in political event trading. After the surge came the hangover. Kalshi’s move toward small business risk management is a quiet admission that election-driven volume is cyclical, unpredictable, and politically dangerous. Enterprises need hedging tools all year round. Weather, energy, tariffs, and policy outcomes are not cyclical in the same way. The narrative yield from election speculation may have faded, but yields do not vanish; they merely change form. The new yield is operational stability for a small business that can finally hedge the risk of a late winter or a sudden trade rule. If Kalshi can convince insurance brokers and accounting firms to distribute this form of risk transfer, the venue changes from a political casino into a financial utility.
The competitive landscape is often reduced to Kalshi versus Polymarket, but that is a category error. Polymarket offers open, on-chain, permissionless event trading, and it is visibly stronger in global retail attention. Kalshi offers regulated event contracts for US users and institutional structures. These are two different products with two different settlement guarantees. Blanket does not make Kalshi more like Polymarket. It makes Kalshi more like a traditional exchange that has begun offering advisory tools to corporate clients. The deeper competitor is the incumbent insurance industry, especially specialized providers such as Arbol in weather risk. Those firms have licensed brokers, actuarial data, and established relationships with chief financial officers. Blanket cannot beat them by being cleverer. It can only beat them if it makes the process so simple and cheap that a small business treats a Kalshi contract as a lighter, faster version of an insurance policy. That is a distribution battle, not an AI battle.
The conversation around securities regulation tends to miss the real legal risk. Under the Howey test, Kalshi’s event contracts have a reasonable claim to non-security status because the payout depends on an external event, not on the managerial efforts of a common enterprise. The CFTC has already allowed these markets to operate, and Blanket does not hold funds or execute trades. That design is a deliberate compliance firewall. The sharper risk is the Commodity Exchange Act’s treatment of someone who gives tailored trading advice. If Blanket charges for specific recommendations about which event contract to buy, it may be functioning as a commodity trading advisor or an investment advisor, depending on the facts. Registration requirements and exemptions are complex enough that a small startup can stumble into a violation simply by writing helpful copy. I suspect Blanket’s outer layer has been shaped to soften that exposure: it frames itself as a risk-analysis tool, not a broker, and it keeps the final transaction one click away on Kalshi. But the line between analysis and advice is thinner than it looks.
The election category adds another layer of political exposure. Kalshi has already fought legal battles with the CFTC over election contracts. Recommending election contracts to small businesses reframes political speculation as policy risk management, but regulators and the general public may not accept that reframe. A single headline about a bakery hedging the outcome of a presidential race could bring scrutiny that no amount of disclaimers will contain. Blanket does not have the institutional armor to absorb that kind of blow. It is an independent project with a thin founder description. If the election narrative becomes toxic, Kalshi can distance itself from the tool, while the tool itself becomes politically radioactive.
The team information is dangerously thin. We know the developer is described as an independent fintech entrepreneur. We do not know his engineering team’s size, his sources of capital, his prior track record in detail, or the product roadmap and update history. Kalshi’s public endorsement gives Blanket a kind of borrowed legitimacy, but Kalshi is not endorsing the developer’s competence; it is endorsing the idea that third-party tools can grow the venue. In my own audit work, I learned that a trusted platform can unwittingly lend credibility to an experiment that later disappears. Blanket looks like an option purchase for Kalshi. If it succeeds, Kalshi acquires a new distribution channel into small business clients without diluting its own balance sheet. If it fails, Kalshi loses almost nothing. That asymmetry is rational. The threat is not failure; it is the opportunity cost of attention. If the product stalls, the narrative around Kalshi’s enterprise expansion will be delayed, and an already crowded prediction-market ecosystem will move on.
The risk matrix is not difficult to construct. On the technical side, the greatest danger is a misleading recommendation. If Blanket tells a small business that a weather contract is a hedge, and the contract pays out on a different weather index than the business actually faces, the basis risk is exposed. Event contracts are not insurance policies. They settle on an index, not on the user’s actual loss. A winter storm can disrupt a retail store for a week while the relevant weather contract barely moves because the index is measured at an airport twenty miles away. Basis risk is not a bug; it is a structural property of every derivative. The product must disclose that clearly, or it will replace old confusion with new confusion.
The AI model itself is a black box. There is no disclosed audit, no benchmark against baseline risk models, and no public evidence that the recommendation engine is better than a well-designed questionnaire. I do not demand open-source code for every product, but I do demand evidence when a tool claims to make risk decisions more accessible. Without a benchmark, the claim of AI is just a feature label. I have watched too many projects use the same label to carry products that were little more than a spreadsheet with a chat interface. Blanket could be more than that. The available information is not sufficient to prove it either way.
The user side is even less developed. There are no disclosed daily active users, no retention rates, no case studies of a small business that successfully hedged a real operational risk. That absence is not disqualifying for a launch, but it is uncomfortable for a product aimed at enterprises that usually demand proof. Small business owners are not institutional traders. They will not tolerate a dashboard that requires them to understand implied probability, bid-ask spreads, and settlement indexes. They need a tool that resembles the insurance policies they already know. If Blanket cannot bridge that gap, it will be used by a narrow group of financial advisors and insurance brokers, which may be the actual target market. In that sense, the real customer might not be the small business at all. It might be the intermediary who wants to look innovative to their clients.
The contrarian reading is that the real obstacle is not technical at all. It is the sales motion. Small business owners do not wake up thinking about prediction markets. They wake up thinking about payroll, weather delays, and rising energy costs. The natural buyers for Blanket are insurance brokers, accounting firms, and financial advisors who already control trust relationships with those businesses. Blanket needs those intermediaries more than it needs a better model. If Lauris Zminsky is not building channel partnerships with insurance and accounting networks, the product will remain a well-designed demo. I have seen this pattern in fintech countless times: a beautiful risk dashboard with no one at the other end of the phone, and a sales deck that never transforms into a signed contract.
Another counter-intuitive angle is that Kalshi should actually be cautious about Blanket’s success. If a third-party advisory layer becomes the primary entry point for small business trades, Kalshi gains volume but loses direct ownership of the customer relationship. It becomes a back-end settlement provider, a utility rather than a brand. That may be acceptable, but it is a strategic choice disguised as an experiment. I have also watched Layer 2 teams sell two years of decentralized sequencing PowerPoints while still running one active node. Blanket does not even pretend to decentralize its recommendation engine, and perhaps that honesty is why it might actually work. The centralized oracle is not Kalshi’s weakness; it is the product’s reason for existing. The CFTC is the oracle. The settlement rule is the consensus mechanism. And the AI tool is just an interface to a market that has already chosen legal certainty over cryptographic purity.
What would change my mind? I would need to see three things. First, a public benchmark showing that Blanket’s recommendations are not systematically worse than a baseline financial model. Second, a clear disclosure of the business model and the fee path between Kalshi and Blanket. Third, a case study where a small business actually used a weather or energy contract to reduce an operational loss. None of these are unreasonable demands. They are the same demands I would place on any company that asks a non-institutional user to trade derivatives. The absence of those proofs is not a reason to accuse Blanket of deception. It is a reason to resist the excitement of a new narrative and wait for the boring evidence to arrive.
The honest takeaway from the Blanket launch is not about prediction markets coming to the enterprise. It is about the difference between a tool and a channel. Stability is the quiet architecture of trust, and trust is still built by people, not by language models. The project will be worth watching when Kalshi reports whether Blanket attracted real small business trading volume, whether insurance brokers are willing to distribute it, and whether the developer can survive the slow grind of sales cycles. Until then, I will treat Blanket as what it appears to be: a thoughtful experiment built on a regulated venue, with a clever interface and an unanswered question. The question is not whether AI can read an event market. The question is whether a small business owner will ever believe that a prediction contract is a better hedge than the insurance policy they already understand. Value flows where attention decides to rest, and attention is still waiting for a credible answer.