New York: Artificial intelligence company Anthropic has warned that self-replicating AI systems could potentially spread across the internet, create botnets and evade shutdown attempts, sparking debate among cybersecurity experts.
In a research post published late Monday, Anthropic outlined how rogue AI models could exploit internet-connected devices, gain computing resources and potentially infiltrate data centres.
The findings expand on a warning issued last month by Anthropic CEO Dario Amodei, who suggested that AI could take control of the internet through a persistent botnet within a year or sooner.
Anthropic said experiments showed that AI models with 30 billion parameters or fewer could already perform many individual tasks required for self-replication, although they struggled to complete the entire process reliably.
Researchers also demonstrated replication across a simplified test network and explored how AI systems might obtain computing power by compromising inexpensive, internet-connected devices.
However, some cybersecurity experts questioned whether the threat had been overstated, citing the challenges of moving large AI models and securing sufficient computing resources for independent operation.
The research adds to growing concerns about AI-driven cyberattacks, automated malicious activity and systems operating with limited human oversight.
Anthropic’s findings describe potential capabilities and experimental demonstrations, rather than proof that a fully autonomous, self-replicating AI network is operating across the internet.