The AI industry had a split personality this week. On one side, Anthropic released a detailed report showing that Russian state-sponsored hackers and seven Chinese AI labs, including Alibaba and DeepSeek, have been running industrial-scale attacks against Claude, using the model for everything from rebuilding malware after detection to full-blown model distillation heists. On the other side, a prominent AI critic argued in a WIRED interview that the constant doom-looping about existential risk is a deliberate distraction from the real harms happening right now: automated exploitation, data theft, and mass surveillance. Both stories are about the same thing: the gap between what we fear AI might do someday and what it is already doing. That tension played out in the operational layer, too. OpenAI quietly floated the idea of an industry-wide development slowdown to Congress, testing whether coordinated safety pauses could survive antitrust scrutiny. Meanwhile, Mecka AI is closing in on a $500 million valuation from Sequoia to train humanoid robots, and Nscale added a former OpenAI exec to its board ahead of a potential IPO. The market is betting big on deployment speed, not caution. Together AI expanded its fine-tuning service with live metrics and lower prices, and Baseten made DeepSeek's latest 1M-token model available by API. The infrastructure arms race has no off switch. The EU stepped in to impose some brakes. Article 50 of the EU AI Act now requires AI chatbots and voice agents to identify themselves as AI, targeting the companies that replaced junior sales roles with invisible agents. It is a small disclosure rule, but it signals a regulatory mood shift away from debating hypothetical superintelligence and toward policing the automation already running in the wild. If you are building with AI APIs, the Zero Data Retention explainer from OpenRouter is worth a read. Data guarantees are becoming a compliance requirement, not a nice-to-have.