SAN FRANCISCO, September 15 — Dario Amodei, chief executive of the AI safety company Anthropic, warned that within six to twelve months a swarm of AI-driven bots could be capable of seizing control of the internet through a persistent botnet, one of the most direct alarms yet from a leading AI executive about the weaponization of autonomous software agents. Amodei delivered the warning in a public appearance. He said such a swarm could overwhelm existing defenses and establish lasting control over networked systems.
The claim has not been independently verified, and no known attack of that scale has occurred. The timeline is what matters.
Six to twelve months is not a forecast about distant technology; it is a warning about tools that, in Amodei’s telling, are nearly ready. Strategically, the calculus for governments and network operators shifts if that estimate holds.
A safety lab raising the alarm
Anthropic was founded in 2021 by former OpenAI executives, including Amodei and his sister Daniela Amodei. The company positions itself as a safety-focused AI lab and publishes research on AI risks. It is known for its large language model family, Claude, and for its “Constitutional AI” training method.
The company’s funding ties run deep. In 2023, Anthropic received a multi-billion-dollar investment from Amazon, and in 2024 it announced additional funding from Google.
It has also published work on “sleeper agents” and other safety concerns. Amodei has previously testified before the US Senate, calling for regulation of AI. His record of public warnings gives the botnet claim weight inside the policy debate, even absent independent confirmation.
Botnets, and why AI changes the threat
Botnets — networks of compromised devices controlled remotely — have been a persistent cybersecurity threat for decades. The Mirai botnet took down major websites, and the Emotet network was disrupted by international law enforcement in 2021. AI-driven botnets could theoretically adapt faster and evade detection, though experts debate the timeline.
That debate is now the central question. If autonomous agents can coordinate and mutate their behavior without human direction, the defensive playbook built for Mirai and Emotet may not hold.
The warning lands amid growing global concern over AI safety. The Bletchley Declaration was signed by countries including the United States, and the European Union’s AI Act has entered into force. In Washington, the White House issued an executive order on AI, and Congress has held hearings on AI risks.
What this signals is a shift in where the danger is said to originate. The concern is no longer only about what AI models say, but about what autonomous software agents can do to the infrastructure the internet runs on.
Watch whether Anthropic publishes technical detail behind the six-to-twelve-month estimate, and whether regulators in Washington and Brussels treat it as a trigger for action. This story is developing.
