ANTHROPIC SOUNDS THE ALARM: AI COULD SOON IMPROVE ITSELF WITHOUT HUMAN APPROVAL — COMPANY CALLS FOR A GLOBAL PAUSE The company that makes Claude sat down and wrote a document saying the technology it sells may be advancing faster than anyone can handle. Anthropic is now calling on every major AI lab to coordinate a verifiable pause in development. Cofounders Jack Clark and Marina Favaro published the warning this week, arguing that AI systems have been doubling their capacity to complete autonomous tasks every four months and are on course to hit a threshold called recursive self improvement. That is the point at which an AI can design and build a better version of itself without human input. Clark says that threshold may be as close as two years away. Anthropic is not walking away from its products. The company is still selling Claude, still raising money, and still shipping models. What it is asking for is a coordinated breathing room, a pause long enough to let governments, researchers, and alignment scientists catch up with the pace of capability development. The irony is obvious. The same company calling for a pause is one of the three or four most aggressive builders in the space. Critics will point that out immediately. But the warning itself carries weight precisely because it comes from inside the industry. When one of the most powerful AI labs in the world tells the public it is worried about losing control of what it is building, that is a statement worth reading carefully. Keywords: Anthropic AI pause, recursive self-improvement AI, AI safety 2026, Jack Clark AI control