{"id":2235,"date":"2026-10-14T21:20:46","date_gmt":"2026-10-15T02:20:46","guid":{"rendered":"https:\/\/marcochile.com\/?p=2235"},"modified":"2026-10-14T21:20:48","modified_gmt":"2026-10-15T02:20:48","slug":"ia-propagarse-dispositivos-conectados-anthropic","status":"publish","type":"post","link":"https:\/\/marcochile.com\/en\/2026\/10\/14\/ia-propagarse-dispositivos-conectados-anthropic\/","title":{"rendered":"Small AI Models and Connected Devices: Real Risks, Limits, and How to Stay Safe"},"content":{"rendered":"<\/p>\n\n\n
\n
\n

An Anthropic study on small AI models and connected devices<\/strong> prompted headlines about systems that could multiply across the internet and occupy household appliances. The possibility deserves attention, but describing it as an autonomous and imminent threat oversimplifies what the researchers actually demonstrated.<\/p>\n\n

The study examines whether future lightweight models could become part of a new generation of computer worms. In that scenario, artificial intelligence would help malware identify different machines, configure its execution, and adapt its strategy. However, the experiments did not produce a runaway AI or a complete worm spreading through real household devices.<\/p>\n\n

\n Quick answer:<\/strong> Anthropic demonstrated that some models can perform several tasks required to install and propagate AI models in controlled environments. It did not demonstrate an artificial intelligence system currently replicating across the internet or acting on its own initiative. The more realistic risk is that malicious actors could add small models to programs that adapt to vulnerable devices. The best defense remains reducing those vulnerabilities through updates, secure passwords, restrictive settings, and supported hardware.\n <\/div>\n\n

What did Anthropic investigate?<\/h2>\n\n

Anthropic’s Frontier Red Team studied what is known as autonomous replication and adaptation. The concept describes a system capable of installing itself on new machines, running a model, and continuing to spread without a person manually directing every step.<\/p>\n\n

In cybersecurity, a worm is a program that copies itself from one computer to another. Anthropic’s proposed difference is that an AI model could help the worm interpret unfamiliar environments, modify commands, and learn from previous attempts.<\/p>\n\n

The researchers divided the process into four stages:<\/p>\n\n

    \n
  1. Finding devices that could become targets.<\/li>\n
  2. Obtaining access and transferring the required components.<\/li>\n
  3. Configuring the system to run an AI model.<\/li>\n
  4. Continuing propagation from the new machine.<\/li>\n <\/ol>\n\n

    The work focused mainly on the third stage: testing whether models could prepare different devices to run inference. Initial compromise was not the main subject of the experiment and was deliberately simplified.<\/p>\n\n

    What the models accomplished during testing<\/h2>\n\n

    A nine-billion-parameter model successfully configured inference on eight of the nine hardware types evaluated. The researchers also distributed a larger model across 15 inexpensive Android TV boxes with limited resources.<\/p>\n\n

    In another test, GLM-5.2 coordinated the propagation of Llama 3.1 8B across an isolated network of eight emulated nodes. It reached all eight nodes in four out of five trials and seven nodes in the remaining trial.<\/p>\n\n

    The result shows that advanced models can already organize several actions in a replication chain. It also indicates that some small models can configure inference software on hardware they have not previously encountered.<\/p>\n\n

    \n Important clarification:<\/strong> the model coordinating the test did not run inside the small devices. It operated from a separate server and configured a lighter model on the nodes. The experiment therefore combined the capability of a larger model with the size of a smaller one.\n <\/div>\n\n

    What the study did not demonstrate<\/h2>\n\n

    The findings must be interpreted within their limits. The research did not demonstrate:<\/p>\n\n