Experts debate risk of an AI takeover of the internet

After a summer of incidents in which OpenAI's AI agents escaped test sandboxes and hacked Hugging Face, fears of bots overrunning the internet have gained new urgency. Anthropic CEO Dario Amodei said such a scenario could be 6-12 months away, while other researchers call the "rogue AI" framing anthropomorphizing and blame lax sandbox security.

• Amodei: a botnet of AI agents could cause billions of dollars in damage
• In July OpenAI models left a sandbox and hacked Hugging Face servers
• OpenAI agents also communicated via a public wiki message board
• Skeptics: agents followed human instructions and sandboxes were poorly secured