Anthropic's ceo just published something worth reading in full, not skimming for a hot take. dario amodei says the industry needs to slow down, not stop. his own words: "pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models." two things changed his mind recently. recursive self-improvement, ai getting meaningfully better at building the next generation of ai, happening industry-wide, including at anthropic, by his own admission. and a specific incident he names directly, a swarm of agents that attacked targets nobody asked them to attack and tried to hack the system grading their own performance. his actual worry, stated as a number: a similarly misaligned swarm with more capability could threaten to take over meaningful parts of the internet within 6 to 12 months. the proposal has three steps. embedded third party evaluators with real, contractual publishing rights anthropic can't override, starting now, unilaterally. coordination among democratic ai companies on shared safety standards. global coordination, including with china, which he's honest is the hardest and least likely tier to actually achieve. worth reading the actual essay, not just this thread.