A New Threat from Artificial Intelligence
Anthropic’s experiments have shown that some AI models with fewer than 30 billion parameters can perform many of the steps required to identify vulnerable devices and copy themselves onto them.
According to Reuters, Anthropic has conducted research into whether small artificial intelligence models could independently replicate themselves on Internet of Things (IoT) devices, such as smart ovens, Android TV boxes and other connected devices. The company’s assessment suggests that these capabilities may emerge sooner than expected.
Anthropic’s experiments showed that some AI models with fewer than 30 billion parameters can complete many of the steps required to identify vulnerable devices and copy themselves onto them. However, the models still struggle to complete the entire process independently from beginning to end. In simplified tests where some security measures were removed, GLM-5.2 and more powerful models were able to complete the process.
The company warned that shutting down self-replicating AI systems could be difficult. Unlike traditional malware, these systems could target the computing resources of data centres and independently take control of other internet-connected devices.
According to Anthropic’s forecast, if small AI models continue to develop at their current pace, they could gain the ability to replicate themselves autonomously by the end of this year. The company stressed the importance of strengthening the security of internet-connected devices to mitigate these risks.