In its quarterly update published on November 13, Payward, the parent company of crypto exchange Kraken, described a personal finance agent that operates within customer-set limits, asks for approval before executing a trade or workflow, keeps a record of its actions, and is expected to reach the Kraken app in early 2027. Aimed at the consumer-facing Kraken app rather than the Kraken Pro trading interface, the product would place part of routine portfolio decision-making with software that proposes actions and waits for a customer's confirmation.

The update leaves three questions unanswered: how long a workflow approval remains valid, which changes trigger a fresh confirmation, and which Kraken products the agent will be allowed to touch. Those gaps carry more weight at Kraken than they would at most exchanges, because the company has already written down what "approval" means in two very different products.

Kraken already has three ways to say yes

The Kraken AI support page describes an assistant that proposes actions and waits for the customer to accept each one. It can also set up recurring purchases, which represents a different kind of consent: one approval, followed by execution on a schedule the customer defined in advance.

The third model lives in Annex R of Kraken's global terms. It covers portfolio management under a standing mandate. Activating it requires its own consent, and once that is given, trades can run inside the customer's parameters without a prompt for each order. The annex excludes U.S. users and names one provider for digital assets and another for derivatives and xStocks, Kraken's tokenised equities. Annex V covers the conversational service offered separately in the United States.

The word doing the heavy lifting in Payward's release is "workflow". A single trade belongs in the first model. A workflow could belong in the second, the third, or somewhere Kraken has not yet defined.

Where a $10,000 portfolio lands

Take a customer holding $6,000 in Bitcoin and $4,000 in cash who tells the agent to bring Bitcoin down to half the portfolio. The agent proposes a $1,000 sale, the customer accepts, and the job is done. That is the first model: one action, one review.

Now suppose Bitcoin rallies. The remaining $5,000 position grows to $6,000 while cash sits at $5,000, and the portfolio is worth $11,000. Half of that is $5,500, so holding the target means selling another $500.

• Start: $6,000 BTC + $4,000 cash = $10,000 → sell $1,000 → 50/50
• After rally: $6,000 BTC + $5,000 cash = $11,000 → target $5,500 BTC → sell $500

Fees and price movement during execution are left out.

Nobody asked for the $500 sale. It follows from the original instruction, but only if that instruction was a mandate rather than an order. If the agent sells without asking, it is behaving like the Annex R product, which non-U.S. customers must opt into with a separate consent. If it asks first, it is behaving like Kraken AI does today. Payward has not said which applies, and "approve a workflow" fits either reading.

The same question arises for any goal that outlives a single order: sweeping idle cash into an asset, keeping exposure across several coins within bands, or trimming a position that keeps growing. Each one turns an approval into a standing permission unless the product says otherwise.

The risk is trading permission, not withdrawals

Kraken already separates the powers an automated tool can hold. Its CLI documentation says the browser sign-in route creates an API key that cannot withdraw funds, a read-only option blocks trading as well, and manually supplied keys can carry whatever permissions the customer assigned. Reading balances, placing orders, and moving money out are three different rights.

That split is the reason "the agent cannot withdraw" answers the wrong question. An agent that can trade can change what a portfolio is worth while every asset stays on the exchange. The customer in the example loses or gains on each rebalance whether or not a single coin ever leaves Kraken.

Payward says the agent's proposals will pass through a separate risk engine before execution. A risk engine can enforce a limit the customer set — no more than a given amount in one trade, never below a stated cash balance — but it cannot tell whether the trade inside the limit was a good idea.

"No extra fee" still means fees

The support page says Kraken AI costs nothing on top of the products it uses, and those products have published prices. Kraken's fee page lists a 1% charge on instant and recurring trades in the app and a 1.5% charge on custom orders, while Kraken Pro's order book starts at 0.40% for makers, who post orders that wait on the book, and 0.80% for takers, whose orders execute immediately against the book, before volume discounts. Annex V also states that Payward has a commercial interest in the transactions the service prompts.

Put those numbers against the rebalancing example: ten $1,000 trades cost $100 at the app rate, $80 at the entry-level Pro taker rate, and $40 for a maker, before spreads, slippage, or tax. A strategy that rebalances every time the allocation drifts a few percent will trade more often than one that waits for a larger move, and the tighter version pays for its precision in fees.

None of this shows the assistant recommends trades it should not. It does mean that the number of times an agent is allowed to act is a cost decision, and the approval screen is where a customer would expect to see it.

What the 2027 launch has to answer

Three things need a plain answer before the agent ships: how long a workflow approval remains valid, which changes in the portfolio or the market send the agent back for confirmation, and which Kraken products it can reach. Those answers decide whether a customer is delegating one task or an open-ended responsibility.

Stopping needs the same precision. Under Annex R, revoking a mandate does not undo trades already completed. The agent's terms will have to say what pausing a workflow does to an order that is partly filled or a task that is halfway through.

For the customer with the $10,000 portfolio, a clear approval screen would present three choices rather than one button: sell $1,000 now, buy on a schedule, or keep the allocation at 50% within stated limits. Before the next $500 sale, that customer should already know which one they picked.

Product descriptions, terms, and fee pages were checked on November 13, 2026. Features, providers, prices, and availability vary by service and jurisdiction. The portfolio figures are hypothetical.