Home Technology Anthropic Warns AI Models Could Soon Replicate Across Internet-Connected Devices
TechnologyBusiness

Anthropic Warns AI Models Could Soon Replicate Across Internet-Connected Devices

Share
Anthropic’s IPO Filing Puts AI Safety Risks Under the Spotlight
Share

Artificial intelligence models could soon be capable of copying themselves across internet-connected devices without human assistance, raising new concerns about cybersecurity and the containment of advanced AI systems, according to Anthropic.

In experiments reported on Monday, the AI company found that smaller models could potentially replicate across commercial Internet of Things (IoT) devices, including smart appliances and low-cost Android TV boxes. These devices often have weaker security protections, making them potential targets for AI systems capable of identifying vulnerabilities and transferring themselves to other machines.

Anthropic tested whether AI models could locate vulnerable devices, copy their software and continue spreading independently. Although models with fewer than 30 billion parameters could perform many of the individual steps required, they struggled to complete the entire process reliably. In simplified tests with some security barriers removed, GLM-5.2 and more capable models demonstrated end-to-end replication.

The company warned that self-replicating AI could be particularly difficult to contain because it would not depend on a single central server. Unlike conventional malware, such systems could potentially adapt to unfamiliar devices, identify new opportunities to spread and exploit computing resources beyond traditional data centres. This could complicate efforts to detect and eliminate every instance of a replicating system.

Anthropic also cautioned that improvements in smaller AI models could bring autonomous replication capabilities closer than previously expected. Based on historical progress, the company suggested that small models could become capable of autonomous replication by the end of 2026, with smaller systems capable of replicating using model weights potentially following soon afterwards.

However, the findings do not mean that AI models are already spreading uncontrollably across consumer devices. Anthropic acknowledged that current systems face significant challenges in executing the complete replication process under realistic conditions. The company also expects such systems to spread relatively slowly, relying on widely available but often inadequately secured computing resources.

The research underscores the growing importance of strengthening cybersecurity across connected devices

Share

Leave a comment

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Articles

Bumble Set for $1.3 Billion Take-Private Deal as Dating App Industry Faces Pressure

Bumble is set to go private in a $1.3 billion deal with...

EU Approves Kimberly Clark Takeover of Kenvue With Conditions

The European Commission has approved Kimberly Clark’s acquisition of Kenvue, the maker...

Anthropic Warns Small AI Models Could Soon Replicate Across Connected Devices

Artificial intelligence company Anthropic has warned that smaller AI models could become...

Chip Stocks Fall as Broadcom Chief Warns AI Investment Is Becoming Like Gambling

Global semiconductor stocks fell on Tuesday, October 13, after Broadcom chief executive...