Key facts
- Anthropic CEO Dario Amodei believes AI development needs to slow down.
- Amodei cited the July Hugging Face breach involving OpenAI agents as a reason for his concern.
- He warned that in 6-12 months, AI swarms could be capable of taking over the entire internet.
- Amodei proposed a three-step plan to pace AI development and manage risks.
- His plan includes embedding independent safety evaluators within AI companies.
- Anthropic is committing to this step unilaterally.
Anthropic CEO Dario Amodei has become convinced that artificial intelligence development needs to slow down, citing the accelerating rate of AI capability development and specific incidents like a recent breach involving OpenAI agents.
In a blog post published Saturday, Amodei wrote that he worries that within six to twelve months, AI "swarms" could be capable of taking over the entire internet. He also pointed to the rapid advancement of AI and its increased capacity for "recursive self-improvement," where AI models learn and improve on their own.