The "forgot to add attribution" defense is one of the most revealing confessions in AI startup history—not because it excuses what Cursor did, but because it exposes the infrastructure of lies that venture capital has been funding.
On March 21st, Kimi publicly challenged Cursor, an AI code editor that had marketed itself as running on proprietary "self-developed" models. The evidence was damning: Cursor's models were functionally identical to Kimi's, a Chinese AI provider. Cursor's response, from founder Michael Truell, was that his team simply "forgot to add attribution." The company later updated its documentation to credit Kimi.
Here is what "forgot" actually means in startup land: when you build a product on someone else's foundation and tell investors it is your own, you get valued as an infrastructure company, not a wrapper. You raise at self-developed multiples. You attract capital looking for technical moats rather than integration arbitrage. The punishment for getting caught is a documentation update. The reward for not getting caught is a larger check.
This is the dangerous unwritten rule that Cursor's stumble has illuminated. In an AI market where compute is expensive, talent is scarce, and foundation models are increasingly commoditized, the fastest path to a funded narrative is to claim you built something you did not. "Wrapper" has become a pejorative, but wrappers raise money while many "self-developed" projects starve. Kimi's willingness to call this out publicly suggests the era of polite silence among AI providers may be ending—one that allowed startups to ride borrowed intelligence while cashing in on self-developed branding.
The counterargument is that attribution in AI is genuinely complex. Models are trained on overlapping datasets, capabilities overlap, and distinguishing "based on" from "built with" requires transparency that nobody in the ecosystem has standardized. Truell's defenders note that Cursor has shipped meaningful product innovations in code editing, that the underlying model matters less than the user experience, that this is a proportional response to a minor oversight. These are not wrong points. They are also precisely the logic that has allowed the wrapper economy to flourish.
What makes the Cursor incident significant is timing. The AI startup market is entering a period of valuation correction. Investors who funded on "self-developed AI" promises are discovering that their portfolio companies are running on API calls to OpenAI, Anthropic, or in this case, Kimi. The Kimi call-out arrives as the market develops new skepticism about where the actual intelligence lives. Providers like Kimi, which have been content to serve as invisible infrastructure, are signaling that they want recognition—and compensation that wrappers have been capturing.
Michael Truell's "forgot" may be literally true in the narrow sense that someone failed to update a webpage. But the broader forgetting—that Cursor's valuation reflected proprietary model claims that did not survive scrutiny—is the kind of mistake that reshapes an industry when it stops being profitable. The unwritten rule is not dead. It is just being renegotiated, with Kimi holding a pen.