Anthropic CEO Publishes Lengthy Post Addressing Safety Concerns, Stresses Need to Stay at the Forefront of Technology

U.S. AI safety firm Anthropic has recently raised concerns about a draft AI safety bill submitted to U.S. lawmakers. The draft sets a high bar for government intervention—authorizing emergency shutdowns of dangerous AI models only in specific scenarios involving biological weapons, cyberattacks, or AI “runaway” events that could cause over 50 fatalities/injuries or more than $1 billion in economic damage. Additionally, certain provisions may weaken state-level regulatory authority. Shortly thereafter, Nathan Calvin, Strategy Lead at AI safety organization Encode AI, conducted a systematic review of Anthropic’s key decisions over the past year: In February, the company revised its previously stringent safety policy. Its earlier commitment—to pause training or deployment “if safety measures lag behind model capability advancement”—was replaced with a new stance that no longer requires unilateral R&D halts at higher-risk stages, citing concerns that competitors could exploit such pauses to achieve technological catch-up.