For years, the existential risk of artificial intelligence remained confined to conferences, laboratories, and specialized communities. Statements from Anthropic researchers on the risks of superhuman systems and the resignation of Jacob Coxon have instead prompted explicit reactions in the US Congress.
Politicians with widely differing views, from Bernie Sanders to Ted Cruz and other Democratic lawmakers, have called for greater attention and, with varying emphasis, action on the safety of advanced systems. Catastrophic estimates remain contested and do not represent a scientific consensus on the probability of human extinction.
From theoretical risk to political responsibility
The important step is not determining today what percentage to assign to extreme scenarios, but recognizing that more autonomous models pose control challenges that cannot be delegated entirely to the companies building them.
The US debate will need to avoid two opposite mistakes: turning hypotheses into certainties or ignoring high-consequence risks simply because they are difficult to quantify. AI policy is thus entering a phase in which uncertainty, too, must be governed.
