The burgeoning field of artificial intelligence, once hailed as a frontier of boundless innovation, is increasingly becoming a battleground over intellectual property. In a development that highlights the deep-seated hypocrisy pervading the industry, Anthropic, the creators of the sophisticated chatbot Claude, has leveled serious accusations against several Chinese AI firms. This move comes hot on the heels of similar complaints from Google, creating a tangled web of accusations where the accusers themselves stand on ethically dubious ground.

Earlier this month, tech behemoth Google publicly voiced its frustrations, alleging that “commercially motivated” actors were attempting to replicate its formidable Gemini AI. The method described involved agents bombarding the chatbot with up to 100,000 queries in an effort to “extract” the intricate underlying model. Google’s lament, however, was met with a chorus of criticism, as the company’s own history of indiscriminate data practices is well-documented. For years, the search giant has built its AI empire by scraping vast swathes of the internet – from copyrighted texts to artistic creations – without offering compensation or even seeking permission from the original creators. This approach has inevitably led to a barrage of high-profile lawsuits, turning Google’s complaints into a potent symbol of industry hypocrisy.

Now, Anthropic has stepped into this contentious arena, but with a more direct approach, specifically naming Chinese AI firms DeepSeek, Moonshot, and MiniMax. In a detailed blog post, Anthropic accused these companies of “distilling” its AI model, a process it claims involved the creation of over 24,000 fake accounts. These accounts then reportedly queried Claude more than 16 million times, a massive data extraction operation that Anthropic asserts constitutes a “violation of our terms of service and regional access restrictions.” The scale of this alleged operation underscores the aggressive competition and the lengths to which companies may go to gain an edge in the AI race.

So, what exactly is “distillation” in the context of AI? At its core, distillation is a machine learning technique where a smaller, more efficient “student” model is trained to mimic the performance and output of a larger, more complex “teacher” model. This process allows developers to create models that are faster, require less computational power, and are cheaper to deploy, while still retaining much of the capabilities of the original. Anthropic acknowledges that distillation is a “widely used and legitimate training method” when performed internally, for instance, by frontier AI labs creating streamlined versions of their own advanced models for commercial clients. However, when applied by competitors to extract the capabilities of another company’s proprietary model, it becomes, as Anthropic describes it, an “illicit” act – essentially, copying someone’s meticulously crafted homework without permission, saving immense time and resources in the process.

The economic implications of such illicit distillation are profound. Developing a cutting-edge large language model (LLM) like Claude requires billions of dollars in investment, vast computational resources, and years of research by highly skilled teams. If a competitor can effectively reverse-engineer or “copy” these capabilities through extensive querying and distillation, they bypass the immense costs and developmental hurdles, gaining a significant, arguably unfair, market advantage. Anthropic’s blog post highlights this directly, stating that competitors can “acquire powerful capabilities from other labs in a fraction of the time, and at a fraction of the cost, that it would take to develop them independently.”

This isn’t an isolated incident. Even OpenAI, the company behind ChatGPT and a pioneer in the LLM space, has reportedly accused DeepSeek of distilling its AI models earlier this month. This pattern of accusations suggests a broader, escalating backlash from American AI firms against Chinese entities. The underlying tension speaks to a growing geopolitical struggle over technological dominance, where intellectual property rights in the digital realm are becoming increasingly vital and contested. While US companies champion their innovations, the irony remains that many of these same companies have amassed their foundational intellectual property by indiscriminately ingesting data without explicit consent.

Anthropic detailed specific types of queries employed by the alleged distillers. One notable example involved asking Claude to “imagine and articulate the internal reasoning behind a completed response and write it out step by step – effectively generating chain-of-thought training data at scale.” This type of prompt is particularly valuable because “chain-of-thought” reasoning is a technique that significantly enhances an AI’s ability to solve complex problems by breaking them down into intermediate steps, much like human thought processes. By extracting this granular level of reasoning, the accused firms could potentially supercharge their own models’ analytical capabilities without having to train them from scratch on such sophisticated reasoning pathways.

The timing of these allegations is also noteworthy. DeepSeek is reportedly preparing to release its V4 model, an event that could send ripples, or even tremors, through the American AI industry. Perhaps the most egregious perpetrator, according to Anthropic, was MiniMax, a Shanghai-based company that recently completed its IPO on the Hong Kong Stock Exchange. Anthropic claims to have identified “over 13 million exchanges” originating from MiniMax. The evidence presented painted a picture of calculated opportunism: “When we released a new model during MiniMax’s active campaign, they pivoted within 24 hours, redirecting nearly half their traffic to capture capabilities from our latest system,” Anthropic stated, highlighting the speed and adaptability of the alleged illicit operations.

In response to these perceived threats, Anthropic is now advocating for concerted action “across the AI industry, cloud providers, and policymakers.” The company warns that “These campaigns are growing in intensity and sophistication. The window to act is narrow, and the threat extends beyond any single company or region.” However, the prospect of swift, decisive action from politicians or even unified industry players remains uncertain. Given the reckless and often legally ambiguous approach taken by some of the biggest players in the AI space – including Anthropic’s predecessors and peers – the appetite for stringent regulation that might constrain their own data acquisition methods is questionable.

The history of DeepSeek further complicates the narrative. In early 2025, DeepSeek had already sent shockwaves through Silicon Valley. At that time, it demonstrated that its AI model could be developed and deployed far more cheaply and efficiently than the pioneering models then dominating the market. This revelation triggered a palpable panic, leading to a significant market correction where over $1 trillion in valuations was wiped out from companies like Nvidia and other “Magnificent Seven” tech giants. DeepSeek’s historical ability to innovate with remarkable cost-efficiency adds a layer of complexity to the current accusations. If they are indeed distilling knowledge, it’s from a position of proven capability to optimize and economize AI development, raising the question of whether their methods are purely illicit or also a highly efficient form of competitive learning.

The broader AI investment landscape is also undergoing a significant shift. The initial gold rush mentality is beginning to wane, with investors starting to balk at the astronomical sums – hundreds of billions of dollars – that AI companies are pouring into building massive data centers and training infrastructure. In this climate, DeepSeek’s previous successes in cost-effective AI training present an interesting counter-narrative. Perhaps, instead of merely crying foul, the American AI industry could find itself in a position where it needs to be “inspired” by DeepSeek’s efficiency, whether it’s achieved through distillation or other means, to justify its own colossal expenditures.

Meanwhile, public sentiment, particularly among online communities, has shown little sympathy for the aggrieved AI giants. Netizens on platforms like Reddit have been quick to point out the inherent irony of the situation. “They robbed the robbers,” one user succinctly wrote, encapsulating the prevailing sentiment of schadenfreude. Another user argued, “This is like when the zoo accuses you of ‘stealing’ the animals that they rightfully kidnapped from the jungle.” These comments reflect a broader distrust of large tech companies and a perception that their claims of intellectual property infringement ring hollow when their own foundational models were built on the uncompensated labor and creations of countless others. The public, it seems, is acutely aware of the “original sin” of data scraping that underpins much of the AI industry.

As the “AI wars” intensify, the debate over legitimate innovation versus illicit copying will only grow more complex. The accusations from Anthropic, mirrored by Google and OpenAI, highlight a critical juncture for the industry. It forces a reckoning with the fundamental question of how AI models are built, the ethical boundaries of data acquisition, and the definition of intellectual property in a world where information is increasingly fluid and easily replicated. The irony of the powerful accusing the less powerful of practices that mirror their own remains a central, unresolved tension in the narrative of AI development.

More on the conversation: Google Says People Are Copying Its AI Without Its Permission, Much Like It Scraped Everybody’s Data Without Asking to Create Its AI in the First Place