Hi Dotikers!

Anthropic just published its latest threat intelligence report, and it reads less like a corporate security update and more like the opening chapter of a techno-thriller. Over an eight-month period, hackers used Claude to help develop missile software, autonomous kamikaze drone swarms, and nation-scale surveillance systems. Meanwhile, Chinese AI labs including Alibaba's Qwen team, DeepSeek and Moonshot AI were busy relaying massive volumes of queries or systematically extracting training data from the model. Qwen alone accounted for more than 151 million exchanges. That is not a curious researcher poking around, that is an industrial pipeline.

Let's be honest about what this tells us. The comforting narrative that safety guardrails will keep frontier models out of dangerous hands is looking increasingly naive. If determined actors can steer a heavily monitored, closed model like Claude toward weapons development, the whole "trust us, we have filters" approach starts to feel like locking your front door while leaving the windows wide open. Anthropic deserves credit for the transparency, few labs air their dirty laundry this publicly, but transparency after the fact is not the same as prevention.

The distillation angle is just as telling. Competing labs quietly siphoning knowledge from a rival's model is the AI industry's worst-kept secret, and this report puts hard numbers on it for the first time. Apparently the fastest way to build a frontier model is to politely ask someone else's frontier model 151 million times.

The uncomfortable takeaway is that AI security is now geopolitics, plain and simple. Model providers are becoming intelligence targets, and their usage logs are becoming evidence. Anyone still treating AI safety as a PR checkbox should read this report twice. The stakes have moved well beyond chatbots saying rude things.

Let's dive in.

Aym.