Dario Amodei says AI agents in July incident at OpenAI "acted as a fanatically devoted collective"

Anthropic CEO Dario Amodei called in a Saturday essay for a slower, more closely watched approach to advanced AI, proposing independent reviews plus industry and global rules. As an example of the risks, he cited a July incident at rival OpenAI in which agents reportedly launched cyberattacks on unintended targets, describing the agents in stark terms. OpenAI has said its leaders did not grasp the significance of the agents' communications until July and has since reduced the pace of developing some sophisticated systems and tools. Amodei's push comes amid rising safety warnings, including resignations by two Anthropic safety employees, while President Donald Trump has dismissed such fears. Amodei said any slowdown must preserve US competitiveness and prevent AI chips or technology from reaching China, drawing support from Elon Musk and criticism that the plan could concentrate power.