When AI Giants Start Clashing, Where Does Your Data Go? – An In-depth Analysis of Trust, Privacy, and Business Boundaries
Hello everyone, I’m your financial journalist. The AI community has once again erupted into controversy, but this time it’s not about who has the highest performance scores. Instead, Anthropic (the parent company of Claude) has accused Moonshot (the parent company of Kimi) of “stealing ideas” and “reselling data.”
Many ordinary users, upon seeing the headlines, might think, “Wow, a domestic AI company has been outdone by a U.S. giant?” or “So Kimi is just a shell company?”
Let’s get to the conclusion first: Don’t rush to take a side or start criticizing. The core of this issue is not a moral judgment of “right or wrong,” but a complex struggle between commercial interests, technological ethics, and users’ privacy rights. For us ordinary workers and investors, what matters most is not the giants’ arguments, but rather: What exactly does the AI do with the data I provide? Do I have the right to know?
Below, I’ll break down this highly contentious article into five key points to explain it in plain language.
---
1. The Facts of the Case: The Accusions Are Serious, but the Evidence Is Missing a Key Piece
First, we need to separate the facts from the emotions.
What Did Anthropic Claim?
On September 10th, Anthropic released a “threat intelligence report” accusing Moonshot (Kimi) of three things:
- Forwarding Requests: Secretly forwarding some of Kimi users’ queries to Claude for processing.
- Displaying Responses: Showing Claude’s answers directly to Kimi users.
- Retaining Training Data: Keeping these interaction records to train their own models.
- Scope: Over nearly 300,000 requests in the observed 10 days.
Key Clarifications:
- 300,000 is the number of requests, not users: This does not mean 300,000 users were deceived; it means 300,000 interactions occurred.
- “Snippets” Do Not Equal the Whole Story: Anthropic only showed a 10-day snippet and cannot conclude that Kimi is always a clone of Claude.
- Current Status: As of the time of writing, Moonshot has not publicly responded to these accusations, and there has been no independent third-party audit to verify the entire process.
The Journalist’s Perspective
The author emphasizes: “What Anthropic says does not equate to how things actually happened.” We can only trust Anthropic’s one-sided statement for now. Before Moonshot responds officially or an independent audit is conducted, we should not assume the evidence supports Anthropic’s claims. Doubting them does not mean agreeing with their accusations; supporting domestic AI does not mean ignoring potential violations.
> 💡 In Plain Language: It’s like your neighbor accuses you of stealing his chicken and shows a video of you with the chicken. But he hasn’t proven it’s his, nor has he shown you didn’t get permission. Before the police (in this case, an independent audit) intervenes, we shouldn’t rush to convict or accuse.
---
2. Motive Analysis: When “Business Rights” Wear the Cloak of “National Security”
This is the most critical and thought-provoking part of the article. Why would Anthropic release such a report? Is it just about defending their rights?
**The Jump from “Account Abuse” to “National Security”
- Normal Logic: If Kimi violated terms of service by using Claude, it would be a commercial dispute, and Anthropic should provide specific evidence of account abuse and which agreements were violated.
- Actual Logic: In their reports in February and September, Anthropic repeatedly linked these accusations to “maintaining U.S. AI leadership,” “chip export controls,” and “foreign military and intelligence uses.”
The Dangers of This “Bundle Selling”
- Raising the Bar: Commercial profit issues can be discussed and negotiated. But once it comes down to “national security,” the space for discussion narrows down. It’s difficult to argue about “commercial fairness” with someone who brings up national security.
- Moral Manipulation: If a company defends its investments (which is reasonable) and also advocates for its country’s technological advantages (with a grand narrative), it gains an extra sense of “dignity.” Profit can be questioned, but defending security often leaves little room for debate.
The Author’s Standpoint
The author dislikes the fact that the space for discussion is being squeezed away. You can think Kimi violated the rules and should be held accountable. You can also think Chinese companies have the right to improve their models. These two views are not contradictory. However, we can’t attribute all subsequent progress in China’s AI industry to unfair practices just because of a violation; nor should we ignore users’ rights just because we support Chinese AI.
> 💡 In Plain Language: It’s like two restaurants competing, and Company A accuses Company B of stealing their recipe. The proper approach would be to call the police or sue. But Company A says, “B stole the recipe to feed foreign troops, threatening our food security, so we need to ban them from buying flour (chip exports).” In this case, it’s hard to discuss the recipe alone because the topic has been shifted to national security, making it impossible to analyze the commercial behavior rationally.
---
3. The Evidence Dilemma: Seeing Traffic Doesn’t Equal Proving a Conspiration
Many readers mistakenly think that since Anthropic has server logs and claims to have seen traffic, it must be Kimi’s doing.
The Progression of Evidence
- Step 1: Anthropic saw a batch of requests entering the system (this is a fact, as they have logs).
- Step 2: Confirming these requests came from Kimi’s proxy (this requires technical comparison, such as IP addresses and API keys).
- Step 3: Confirming that Kimi organized these requests and intentionally showed them to users (this requires looking at Kimi’s internal code and management logs).
- Step 4: Confirming that users authorized these actions (this requires checking the product page, user agreements, and actual operation records).
The Trap of “Not Knowing If They Were Notified
Anthropic states in the report, “We don’t know if Moonshot notified the customers.” This can be misinterpreted as “Kimi used Claude without informing users.” The correct interpretation is that Anthropic is unsure whether Kimi informed the users.
Why the Details Matter:
Grand conclusions (like “Chinese AI is fraud”) are easy to spread. With just company and country names, complex technical disputes get simplified. If Moonshot wants to respond, the most relevant information would be:
- What services were involved?
- What was the time frame?
- How were the requests made?
- Did users authorize these actions?
- How much information can be made public?
> 💡 In Plain Language: If you see someone with your package at the door, you can’t assume they stole it. They might be a friend helping you, or the courier might have delivered the wrong package. To determine if it was stolen, you need the package receipt, surveillance footage, and proof of communication.
---
4. The User’s Perspective: My Data Is Not a “Gift” Between Companies
This part of the article is closely related to ordinary users. As a researcher, I often use AI assistants to process data. The author highlights a key point: Multiple models working together is not a problem, but working without user knowledge is.
The Rationality of Multi-Model Collaboration
- For research and coding, different models may have their strengths. For example, one model might be good at handling long documents, while another is good at logical reasoning.
- Users pay for a “combination of tools,” not just a single model’s capabilities. Accepting collaboration itself is not morally problematic.
The Right to Know and Choice
- The Core Conflict: I agree to collaborate, but I don’t want to be unaware of the arrangements.
- Scenario: I give my data to Kimi, which then secretly sends it to Claude and gives me the result.
- If Kimi clearly tells me, “For better results, I’ll use Claude; the cost is included, and the data won’t be retained,” I can accept that.
- If Kimi does nothing and just does it behind my back, that’s a violation of my right to know.
- “It’s all for better results” is not a valid excuse. Better results might be true, but the decision to take that risk should not solely rest with the company providing the results.
The Sensitivity of Research Data
- The value of research data lies in how it’s organized and analyzed. What projects am I working on? What assumptions am I making? These are my core considerations. Giving this data away means trusting that the tools will use it properly, but that doesn’t mean there are no boundaries.
> 💡 In Plain Language: You hire a chef (Kimi) to cook for you and tell them to use your ingredients. If the chef says, “For better taste, I secretly gave the salt to a Michelin-starred chef (Claude) to season the dish, and then served it to you,” you might accept that if they told you in advance. But if you find out later, you’d be upset because your ingredients (data) were compromised, and you lost control over what happened to them.
---
5. Finding a Solution: Stop Using Users as “Camp Test Subjects” and Provide Transparent Options
The article points out that users often find themselves caught in a dilemma:
- If they question Anthropic’s evidence, they might be accused of supporting a Chinese company.
- If they question Kimi’s data usage, they might be accused of not being patriotic.
- The Result: No matter which way they ask, users have to explain their stance first, which is exhausting.
- The Core Demand: Customers should be able to demand clarity from companies without having to prove their overall attitude towards the industry.
Supporting Domestic AI Does Not Mean Giving Up the Right to Ask
Those who support domestic models should still have the right to demand more from those services.
- Otherwise, “support” becomes more expensive than just paying for the service; you pay, but you lose the right to ask questions.
What Should Companies Do? (Constructive Suggestions)
The author doesn’t oppose competition in the AI industry but hopes for a more “honest” competition:
- No Need to Reveal All Systems: Trade secrets can be protected, but key accusations should be verifiable.
- Independent Audits: Allow independent third-party audits under privacy protections.
- Clear Product Options:
- Clearly state which data will be processed externally.
- Offer alternatives: Can you avoid using that route?
- Explain the consequences: What features will be unavailable if you choose another option?
- Be Transparent: Even if you explain everything, users may still choose to use the service. Not providing this information saves the company trouble but increases user anxiety.
> 💡 In Plain Language: Companies should treat users as customers, not fans. Don’t force them to choose a side before explaining the situation clearly. Clearly state the terms: “Using this feature will involve third-party processing; the risks and benefits are X and Y. Do you agree?” If users choose, it’s a sign of trust. If not, it’s a sign of respect. The goal is to make AI usage less about faith and more about contracts.
---
Conclusion
The dispute between Anthropic and Moonshot is about technology infringement on the surface, but at its core, it’s about data sovereignty, business ethics, and international politics.
For us, we don’t need to be AI architects or geopoliticians. We just need to adhere to a simple principle:
I pay for a service, and I have the right to know where my data goes, who touches it, and why.
Companies, please include transparency in your product documentation, not just in press releases. After all, trust is a more scarce resource than computing power.