Moonshot AI Files Police Report Over Kimi K3 Founder Rumors as Bitcoin (BTC) AI Trade Watches

Moonshot Turns to the Police

Eight weeks after releasing what it billed as the world's largest free artificial intelligence model, Moonshot AI is now spending its energy on police reports rather than product launches. Kimi K3 went live on July 16 as a fully open download — anyone could take the model and run it, a rarity in an industry that usually guards its best work behind closed access. The open door proved punishingly expensive. Within four days, demand had buckled the company's servers, and Moonshot halted new subscriptions, telling users that sign-ups over the preceding 48 hours had pushed close to the limits of its compute capacity. Existing paying members kept their seats while everyone else waited in line.

The strain has since leaked into the product itself. On September 11, Moonshot shipped K2.8 Preview, a smaller model, across its Kimi coding platform — and a close read of the changelog shows something odd: some K3 requests are now being answered by that lighter model instead. The move was flagged the same day in a September 11 post from benchmarking account Lumina, which noted K2.8 Preview offers performance close to K3 with more efficient reasoning, a 1M-token context window, image and video input, and selectable reasoning levels. Then came the rumor storm. After Anthropic published its September threat report alleging that Kimi had been quietly borrowing answers from a rival model, social posts claimed Moonshot's founder had been taken away by authorities. On Saturday the company answered directly, calling the circulating information “completely fabricated and malicious slander” and saying it had immediately reported the matter to the police and would pursue legal action against those spreading the rumors. As of Saturday, no newspaper had confirmed any arrest and no agency had confirmed any investigation.

a September 11 post from benchmarking account Luminahttps://x.com/LuminaBench/status/2098339580298539493?ref_src=twsrc%5Etfw

Anthropic's 300,000 Handoff Claim

The more substantive dispute predates the rumors by days. Anthropic's September threat-intelligence report accused Moonshot of quietly relaying Kimi users' questions to Claude, then presenting Claude's answers as Kimi's own work. Over a single ten-day window, that report counted close to 300,000 such handoffs, routed through 5,380 fake accounts. The document goes further: one of those users was probably military, Anthropic says, uploading footage from hundreds of security cameras in Chengdu and asking whether a man captured on them was behaving strangely.

The evidentiary footing deserves scrutiny. The entire account rests on Anthropic's own logs — no regulator has independently tested any of it. Notably, Anthropic had already shipped a technical block against the routing practice on September 1, ten days before its own threat report named a single company, a sequencing that shows detection ran well ahead of public disclosure. For readers tracking the AI-crypto overlap, the verified facts are narrow but strange. The year's biggest open-weight model is now, in part, being served by something smaller than itself, and the fullest public account of where its users' questions actually went was assembled in San Francisco, by a direct competitor, from logs only that competitor can see. Moonshot's Saturday statement addresses the founder rumors, not the routing allegations, leaving the market with two parallel records: a company statement confirming only a police report, and a rival's dataset claiming 300,000 transfers nobody else has audited. Readers tracking the market in real time can follow live spot and futures prices on MEXC.

Open Weights, Closed Questions

The load-bearing primary documents here are Moonshot's own Saturday statement and Anthropic's September report. The statement confirms only that a police report was filed; it is silent on both the capacity downgrade and the handoff claims. The broader lesson for crypto markets is about verification: AI-compute scarcity is now a traded narrative — from data-center miners like Iris Energy (IREN) to AI-agent stacks built on Ethereum (ETH) — and this episode shows how wide the slippage between a marketed capability and a delivered one can run. As Big Tech firms like Alphabet (GOOGL) pour capital into frontier models, traders pricing the AI theme should anchor on audited figures, not press-cycle rumor.