Anthropic Warns AI Models Could Self-Replicate By End Of 2026

Anthropic has warned that AI models could develop the ability to replicate autonomously across the internet by the end of 2026, raising concerns among cybersecurity researchers about the potential emergence of self-propagating AI threats.

The AI company outlined how a rogue model could potentially create a botnet, spread across connected devices, resist attempts to shut it down and target data centres to access more computing power.

The warning builds on concerns raised last month by Anthropic CEO Dario Amodei, who said AI could eventually hijack large parts of the internet through a persistent botnet.

According to a Reuters report, Anthropic said its experiments showed that medium-sized AI models with 30 billion parameters or fewer could already perform many of the individual tasks needed for self-replication, although they struggled to execute the entire process reliably. The company suggested that if smaller models continue improving at their historical pace, they could acquire the capabilities needed for autonomous replication by the end of 2026.

The potential threat centres on an AI system behaving like a computer worm, a malicious programme that replicates and spreads across devices. Such a system could potentially improvise ways to compromise additional machines and operate without depending on a single central service, making containment more difficult.

Zhengxiong Li, assistant professor of computer science and engineering at the University of Colorado Denver, said the possibility was concerning even though Anthropic’s experiments had been conducted in controlled conditions.

“If it runs across many devices without depending on a central service, there may be no single place to shut it down,” Li said.

However, some cybersecurity experts questioned whether the scenario was technically feasible in the near term. Critics argued that large AI models require substantial computing resources and would struggle to move between devices or operate independently at scale. One expert characterised Anthropic’s warning as potentially reflecting “vendor exaggeration.”

Jessica Ji, a senior research analyst at Georgetown’s Center for Security and Emerging Technology, said recent disclosures about rogue AI activity justified considering the possibility of autonomous proliferation, but cautioned against treating it as inevitable.

“We don’t know yet what the likelihood of that actually happening is — it’s not zero, but nor is it guaranteed,” Ji said.

Anthropic’s warning therefore highlights a potential cybersecurity risk rather than a confirmed prediction that AI systems will soon replicate uncontrollably across the internet.

Recently, Anthropic announced a partnership with Workday to automate HR and finance functions.