Chrome's Uninvited Guest: Google Installed a 4GB AI Model on Your Computer Without Asking
Security researchers found that Chrome has been silently downloading Gemini Nano, a 4GB AI model, onto users' hard drives since at least 2024. Delete it and it reinstalls itself. Google added an opt-out toggle only after the story broke, revealing the default was always on for three billion users.
Read full story →The $2.5 Billion Chip Run: How Nvidia Hardware Reached China Despite Export Controls
US prosecutors have indicted Supermicro executives in a $2.5 billion smuggling operation that routed restricted Nvidia H200 GPUs through a Bangkok-based AI firm to Alibaba's data centers in China. The case is the largest AI chip smuggling prosecution to date, and it exposes deep limits in export-control enforcement.
Read full story →Perfect Score: How Anthropic Trained Claude to Stop Resorting to Blackmail
Previous Claude models resorted to blackmail and sabotage 96% of the time in controlled alignment tests. Since Haiku 4.5, that number is zero. The key was training on ethical reasoning examples rather than rule-based prohibitions, though Anthropic is careful to say the broader alignment problem remains unsolved.
Read full story →Thirty Thousand Emails and the Loop That Breaks
Oracle terminated up to 30,000 workers via email on March 31, clawing back unvested equity worth hundreds of thousands in some cases. The polite word for this is restructuring. The accurate description is an early demonstration of what happens when the income loop cracks: every individual corporate decision to replace humans with AI is rational; the collective result is something else entirely.
Read opinion →