Anthropic’s CEO Jokes About Bunkers—and AI Risks

AGI: Too Powerful to Release Casually

In a candid and eyebrow-raising conversation, Anthropic CEO Dario Amodei said his company is deeply considering the real-world impacts of releasing artificial general intelligence (AGI). At the Fortune Brainstorm AI conference, Amodei described scenarios where AGI could be so powerful—and potentially dangerous—that its deployment would demand extreme safety measures. Half joking, half serious, he remarked, “We’re definitely going to build a bunker before we release AGI,” highlighting the gravity with which Anthropic views the potential consequences of unleashing a truly general AI system. His remarks underscore a broader unease in the AI community about balancing technological advancement with existential safety risks.

Between Safety and Speed

Amodei emphasized Anthropic’s commitment to technical alignment—ensuring that AI systems act in accordance with human intentions—as key to safely developing AGI. While some companies are racing ahead with increasingly powerful models, Anthropic’s approach centers on controlled progress and early infrastructure for containment. The CEO acknowledged that the exact trajectory toward AGI remains uncertain, but stressed readiness for unexpected developments. His comments reflect a growing split in the AI world: one camp driven by innovation pace, the other by caution and governance. If AGI does emerge sooner than expected, Anthropic wants to be prepared for the profound implications—bunkers included.