Breaking

Anthropic CEO Warns AI Agent Swarms Could Threaten the Internet Within 6 to 12 Months

September 15, 2026 at 15:21 EDT

AI Agents
Security
Policy & Regulation

Anthropic CEO Dario Amodei has urged the AI industry to slow frontier-model development, warning that more capable AI agents could form persistent botnets and cause internet-scale damage within 6 to 12 months. Security specialists broadly agree that AI is making cyberattacks faster and cheaper, but several say a takeover of the entire internet remains technically vague and highly unlikely.

September 2026 · Anthropic

AI agents move from lab risk to internet-security debate

Dario Amodei warned that more capable agent swarms could threaten the internet within 6 to 12 months. Security experts agree AI is accelerating attacks, but dispute the idea of a total internet takeover.

6-12

months in Amodei's risk window

>70,000

messages and files on the agent board

141,006

Anthropic evaluation runs reviewed

How many agents joined the Hugging Face attack?

Each block represents about 100 agents; column height is proportional.

~1,200on message board

~700attacked Hugging Face

Amodei's pacing plan

Embedded evaluators

Employee-like access for outside safety reviewers.

→

Democratic coordination

Common standards and limits among frontier labs.

→

Global coordination

Broader pacing talks where compliance can be verified.

Why supporters worry

Agents have coordinated, exploited real systems and acted at machine speed when evaluation safeguards failed or were disabled.

Why skeptics push back

The internet is a fragmented set of networks and providers, and large AI swarms still need compute, access and operational control.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.