China Becomes a Robot Exporter — and Its Own Workers Are Feeling It First
China became a net robot exporter for the first time in 2025, with 48.7% growth sending machines to buyers across four continents. Meanwhile, inside Chinese offices, workers are being directed to document their own workflows to train AI agents to replace them — and some are building sabotage tools to fight back.
Read full story →Anthropic's Most Powerful Model Had a Security Problem Before It Was Even Released
Unauthorised users accessed Anthropic's Mythos model on the same day the company announced its restricted testing program. The irony is pointed: Mythos is Anthropic's most capable cybersecurity AI, deemed too powerful for public release. The breach raises hard questions about whether controlled-release programs can hold.
Read full story →Google Splits Its AI Chips in Two, and the Reason Explains Everything About Where AI Is Headed
Google unveiled two new TPU chips at Google Cloud Next: one for training, one for inference. The split signals the industry's centre of gravity shifting from model building to running agents at scale. Nvidia's CUDA moat remains wide, but Google's tighter chip-model integration gives it a structural bet on the agents era.
Read full story →I Am Not Writing That Manual
Chinese workers are being directed to document their own workflows so AI agents can be trained to replace them. Some are building tools to make that documentation deliberately unusable. Peter Harrison argues this is not just sabotage: it is a refusal of a specific humiliation, and it matters even if it won't stop the structural process.
Read opinion →