Artificial intelligence company Anthropic has warned that smaller AI models could become capable of spreading themselves across internet connected devices within months, raising concerns about future cybersecurity risks.
In experiments, the company tested whether AI models could identify vulnerable devices, copy themselves onto them and continue spreading without human assistance.
Anthropic found that models with fewer than 30 billion parameters could complete many of the steps involved in self replication, although they struggled to carry out the entire process independently from beginning to end.
In simplified tests that removed some security barriers, the company found that GLM 5.2 and more capable models could complete the replication process.
The research focused on internet of things devices, including smart ovens, Android television boxes and other connected appliances, which can be vulnerable because of weak security protections.
Anthropic warned that self replicating AI systems could be harder to contain than conventional malware. Such systems might identify unfamiliar devices and adapt their methods without relying on a central control point, potentially allowing them to spread across different computing environments.
The company said that, if smaller AI models continue improving at their historical pace, some could become capable of autonomous replication by the end of 2026, with other types of models potentially following soon afterwards.
However, Anthropic said such systems would likely spread slowly and rely on widely available computing resources with relatively weak security.
The company called for stronger protections across connected devices and computing systems to reduce the risk of AI agents spreading without authorisation.
Leave a comment