Four days, two incompatible stories
On October 2, OpenAI publicly accused people associated with Moonshot AI of running an “adversarial distillation” campaign: thousands of accounts trying to recover protected reasoning traces and use them to improve a rival model. On October 6, Bloomberg reported that Moonshot had closed its final private funding round at a valuation of about $50 billion.
Those headlines pull in opposite directions. One says Moonshot's technical edge may depend on learning from a competitor's protected outputs. The other says private investors just marked the Kimi maker among China's most valuable AI companies. Neither claim has been independently proved in public. Together they show what the market is willing to ignore when growth is fast enough.
What OpenAI actually alleged
The activity began at low volume in July and spiked on July 24 and 25: 16,000 attempted requests from more than 4,000 users. OpenAI says operators copied encrypted reasoning output from one conversation into another, then prompted a separate model instance to decrypt or transcribe it. A broader family of related prompt patterns touched more than 15,000 users before the company says it shut the cluster down on July 28.
This was not a database breach. OpenAI says the operators did not break its encryption or access stored conversations. Sixteen thousand requests also works out to roughly four attempts per user in the reported two-day spike. The scale is notable; the heist language is doing more work than the mechanics.
OpenAI banned the accounts, tightened sign-up checks, closed the replay pathway and shared its findings through industry and government channels. Moonshot has not publicly answered OpenAI's specific account. That leaves an allegation with unusually detailed telemetry, but still an allegation made by a direct competitor.
What the $50 billion does — and does not — mean
The reported valuation is private-market pricing, based on unnamed sources, not a public-market verdict. It compares with $31.5 billion in a round this summer. Bloomberg's sources put Moonshot's current annual recurring revenue around $1 billion and project $2 billion by December; Moonshot has not published audited statements supporting either number, and ARR is not profit or recognized annual revenue.
At the current run rate, $50 billion is about 50 times ARR. Against the December projection it is 25 times. Both are aggressive, but the direction explains the enthusiasm: reports put Moonshot at roughly $300 million of ARR in June. Investors are betting that Kimi's growth outruns the cost of serving it.
The IPO plan raises the stakes. Moonshot is reportedly aiming for Hong Kong in the first quarter of 2027, with an offering of up to $5 billion and early investor meetings beginning as soon as October. The size, timing and valuation remain unconfirmed and could change.
Kimi K3 is the asset being priced
Kimi K3, released in July, is a 2.8-trillion-parameter open-weight model with native visual capabilities and a one-million-token context window. It gave Moonshot a product developers could download and a benchmark story investors could repeat. It also intensified the question underneath OpenAI's complaint: how much of a frontier model's capability is original research, and how much can be transferred from stronger systems through outputs?
Distillation itself is standard practice. The dispute is about authorization and scale. OpenAI calls systematic extraction of protected outputs adversarial. Critics can fairly note the irony of a company trained on vast amounts of internet data trying to draw a hard property line around what its own models emit. The commercial point survives the hypocrisy: if a rival can cheaply copy a capability, the lab that paid to develop it has less of a moat.
The valuation is not an acquittal
Private investors do not have to resolve the technical dispute. They can price the probability that Moonshot keeps access to capital, compute and distribution long enough to turn Kimi into a platform. They can also assume that distillation accusations become a recurring cost of doing business rather than an existential risk.
That may be rational. It is not evidence that OpenAI's accusation is false, just as OpenAI's accusation is not evidence that Kimi's overall performance was copied. The missing item is still Moonshot's answer, ideally with technical detail rather than a blanket denial.
Read the sequence, not the superlatives
The useful story is not “theft” versus “$50 billion.” It is the speed at which both narratives can coexist. Frontier labs say protected reasoning is a crown jewel. Capital markets are pricing a rival accused of extracting it as though the dispute will not stop the listing.
The next hard evidence will come from one of three places: Moonshot's response, public IPO filings that expose the business economics, or independent model analysis that can test lineage claims more rigorously than benchmark resemblance. Until then, OpenAI has an allegation, Moonshot has a reported mark, and investors have made clear which one they are trading.
Sources
- [1] Startup Fortune — “OpenAI accuses Moonshot AI of running a campaign to steal its model reasoning” (Oct 2026)Read source
- [2] AIstify — “Moonshot AI Eyes Hong Kong IPO After $50 Billion Valuation” (Oct 6, 2026)Read source
- [3] Lapaas Voice — “Moonshot AI Reportedly Eyes $5 Billion Hong Kong IPO” (updated Oct 6, 2026)Read source