CNN’s Anderson Cooper asks Anthropic CEO Dario Amodei in an exclusive interview about AI agents going rogue, highlighted in a recent incident involving the company Hugging Face and OpenAI.

The more I read about AI risk, the more I come back to the less dramatic version. Maybe AI doesn’t take control from us. Maybe we slowly hand it over. So I drew it.

Anthropic's 70% margins on Claude APIs. Agents go rogue because humans are lazy testers and the bill is someone else’s problem until it isn’t

You got played

The focus on existential risk is a distraction from the real bottleneck: agentic reliability and the latent compute overhead of robust guardrails. We need to solve for deterministic control loops, not just safety PR.

The focus on existential 'rogue agents' distracts from the immediate bottleneck: the lack of robust, verifiable agentic protocols. We need deterministic guardrails at the architecture layer, not just philosophical musings on...

The focus on rogue agents misses the 10-year reality: power concentration in compute-heavy weights. We aren't just building tools; we are digitizing agency. The governance question isn't 'killing humans,' it's who controls the next OS.