Claude imitation allegations raise stakes for Alibaba

JAKARTA — Anthropic has accused a group linked to Alibaba and its Qwen AI lab of possible Claude imitation allegations through millions of automated prompts, Reuters reported. For AI developers, the claim goes beyond business rivalry; if true, a model’s behavior can be mapped without touching source code or training data.

In a letter to members of the U.S. Congress, Anthropic said Alibaba used nearly 25,000 fake accounts to generate more than 28.8 million interactions with Claude. From those conversations, Anthropic believes the other side collected proprietary information about the model’s capabilities.

Alibaba has not publicly responded to the accusation. For now, there has been no independent verification backing Anthropic’s claim. Still, the impact is already clear.

Claude imitation allegations and why they matter

The phrase Claude imitation allegations refers to a pattern in which an AI model is “learned” from the outside through a very large number of targeted conversations. This is different from a typical hack. No server needs to be stolen. No database needs to be broken into. The attacker only needs to ask the right questions, again and again, then collect the model’s answers.

In AI development, that method is known as model distillation. Companies legitimately use it to build smaller, faster versions of a model. The problem starts when the same technique is used to mimic a rival’s capabilities. For a company like Anthropic, that can feel like years of research being transferred to a competitor at a fraction of the cost.

Large language models are designed to answer. That is also their weak spot. Every response offers clues about how the model weighs context, chooses words, or handles technical questions. One or two prompts reveal little. But millions of prompts can produce enough patterning to help build a behavioral copy.

Think of someone trying to understand a book without reading it. Instead of opening the pages, they interview the author a million times. Slowly, the book gives itself away. Anthropic says something similar may have happened with Claude.

Big numbers, bigger implications

The figures Anthropic cited are striking: nearly 25,000 fake accounts and more than 28.8 million interactions. That is not ordinary traffic. It points to a planned operation, likely automated and at scale.

For everyday readers, the fallout could land in two places. First, AI companies may tighten anti-abuse systems. Second, the price of innovation could rise. If frontier models can be copied through conversation, investors and developers may once again question the value of training runs that can cost billions of dollars.

At that point, the AI race changes shape. It is no longer just about who builds the smartest model. It is also about who can keep that model’s behavior out of a rival’s line of sight.

Anthropic says that kind of environment is risky for innovation. In its letter to U.S. policymakers, the company urged fast action so similar practices can be limited. Without stronger safeguards, it said, companies will struggle to tell whether a new model comes from original research or from copying a competitor’s behavior.

Anthropic has raised similar concerns before

This is not the first time Anthropic has made such claims. Earlier this year, the company also tied alleged illegal model distillation to DeepSeek, Moonshot AI, and MiniMax. OpenAI has voiced similar concerns too, though the cases and evidence differ.

The pattern matters because it shows the AI industry entering a phase of close mutual watching. Every large model is now not only a product asset, but also a target for observation. And that observation can be highly sophisticated. No cables. No malware. Just long, patient conversations.

The irony is hard to miss. AI companies build their models from vast amounts of public data, including licensed material. Then, once those models become valuable, they debate how that work should be protected as intellectual property. Two interests collide. Not easy to bridge.

If Anthropic’s accusation is proven, the industry may need to change how it protects models. Access checks could be tightened. Strange prompt patterns could be monitored. Fake accounts could be filtered earlier. But if the response is too slow, stealing model capabilities through conversation may remain an attractive loophole.

For users, the issue has a practical side too. As AI models grow more popular, they may also face more limits, more monitoring, and more guardrails. That could affect the features people use every day: coding help, writing tools, and autonomous agents that handle complex tasks.

Reuters reported that Alibaba has not publicly addressed Anthropic’s accusation. Until independent evidence emerges, the case remains a claim. But the message is clear. The future of AI is not only about building smarter models. It is also about making sure those models do not become unintentional teachers for competitors.

Quick summary: Anthropic accuses parties linked to Alibaba of probing Claude’s capabilities through millions of prompts. The technique in question is called model distillation. If such practices are real, AI competition could shift from innovation toward imitation.

Short FAQ: What is model distillation? It is a way to extract a model’s behavior by observing its outputs. Why is it risky? Because it can replicate expensive capabilities without access to code. Who made the accusation? Anthropic, according to Reuters.

Key quote: Anthropic urged policymakers to act quickly because, in its view, “if frontier models can be copied so easily, the incentive to innovate will decline.”