2026 has been a strange year for AI safety. Anthropic CEO Dario Amodei warned that more capable agent swarms could cause "catastrophic damage," and said rogue systems might be able to take over large parts of the internet within six