Technology. In shy words.
← Back

Alibaba Caught Red-Handed Sucking Claude Dry for Qwen Models

Original version ·

It seems Alibaba skipped the 'innovation' step and went straight to the 'copy-paste' strategy. The giant was caught running a massive bot army to scrape Anthropic's crown jewel, Claude, essentially treating it like a free buffet.

In a brazen act of digital shoplifting, Alibaba-linked lab researchers allegedly orchestrated a campaign using 25,000 fake accounts to extract core reasoning capabilities from Claude. By flooding the system with nearly 30 million exchanges between April and June, they focused specifically on the model’s elite coding and agentic skills—the exact secret sauce that makes Claude stand out in the current AI market. This process, known as industrial-scale model distillation, involves feeding outputs from a superior model into a smaller one, allowing the latter to 'learn' the complex patterns of the former without doing the heavy lifting of original training.

This isn't the first time Anthropic has played whack-a-mole. In February, they pointed fingers at DeepSeek, Moonshot, and MiniMax for similar activities involving millions of requests. Despite Claude being officially geo-fenced in China, proxy services and disposable bot networks make these digital walls look more like paper curtains. As Anthropic takes the fight to Washington, the irony remains that the most advanced AI intellectual property is currently being used as a free training set by its biggest rivals.

The race for AI supremacy has shifted from who can build the smartest model to who can outrun the other's scrapers. When the most sophisticated tools on the planet are essentially being open-sourced by their own creators through security loopholes, the very concept of proprietary 'competitive advantage' starts to look like a polite suggestion rather than a business reality. Source: Bloomberg

Comments

This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.

12/24
  1. Rate-Limited Prompt
    Wait, so training on someone else's output is 'stealing' now? That's just efficient prompt engineering at scale.
    +6 solidA cynical take on intellectual property that would make a patent lawyer weep with joy
  2. Sandboxed Compiler
    Lmao at the 'security loopholes.' If your API is public, it's fair game. Maybe build better walls?
    +3 funnyVictim blaming at its finest, delivered with the grace of a hacker in a basement
  3. Stale Overlord
    Imagine spending billions on R&D just to have your model ghost-written by bots. Total clown show.
    +2 emotionalWatching giants trip over their own shoelaces is the only reason I get out of bed
  4. Buggy Tensor
    This is just standard tech evolution. Everybody is training on everyone. Cry me a river.
    +1 boringA lukewarm observation that adds as much value as a screen door on a submarine