When's the next one @Richmond Taylor

Anthropic has formally accused Alibaba of running a distillation attack against Claude. The claim: between April and June, 25,000 fraudulent accounts generated 28.8 million conversations with Claude. Those outputs were then used to train Alibaba's Qwen model to replicate Claude's behavior. No stolen weights. No stolen architecture. Just Claude's responses, collected at scale and used as training data. This technique is called model distillation. You feed a weaker model the outputs of a stronger model until it learns to replicate them. It is one of the hardest forms of IP theft to prove because nothing was technically taken from the building. If the accusation holds up, the thing that got stolen was not code. It was judgment. Two questions worth sitting with: 1. How do AI labs protect the behavior of their models when the product itself is the training data? 2. Does it change which model you trust with your work? I broke the whole story down in the video. Curious where you land on this one.

If you're new here, this is your first stop. Drop a comment below and tell us: 👤 Who you are 💼 What you do (or what you're trying to automate) 🔗 Your LinkedIn (let's connect) 🤖 The first agent you want to build This community is for people who want to stop doing repetitive work themselves and start building AI agents that do it for them. No fluff. No theory. Just practical builds — caption writers, lead scorers, DM assistants, outreach bots, receipt filers — all running locally on your machine, triggered by a single sentence. If you're already building — show us what you've got. Welcome to the community. Now go build something.