The Small Bot Can Be the Bigger Risk
America's bank regulator is telling examiners to judge AI by its authority and the consequences of failure, not by how impressive the model sounds.
Read the pick: The Small Bot Can Be the Bigger RiskX OPPORTUNITY RADAR / SHORT TAKE
Moonshot is resetting Kimi subscriber quotas after unnecessary subagent calls and long outputs consumed usage.

Moonshot AI says problems around the Kimi K3.1 Pro launch consumed subscriber quotas through unnecessary subagent calls and excessively long outputs. The company apologized and said it would reset every subscriber’s monthly quota by 25 percent, which it described as one week of use. Accounts that had used less than 25 percent would return to zero usage.[1]
The company also says it adjusted the subscription plan’s parameters to reduce needless delegation and output length. That is the technical fix. The commercial issue is more interesting: users bought a quantity of service, while the system decided how much hidden orchestration to spend on their behalf.[1]
Reuters also reports that Moonshot has faced compute-capacity pressure, in a Chinese AI sector constrained by access to advanced chips. Capacity pressure may explain strict quotas. It does not settle who pays when the product’s internal agent tree burns through them.[1]
A quota reset is the right first response. It also creates a useful precedent. If an AI service chooses the subagents, retries and token budget, the provider should own the cost of waste the customer could neither see nor prevent. When should reimbursement be automatic rather than an apology after launch day?
Correction ·
Clarified that the reported adjustments concern the subscription plan’s parameters; the reporting does not establish changes to model weights.
Checked Oct 26, 2026. Numbered references point to the evidence behind factual passages. Commentary is our own.