Anthropic warns AI agents could spread like worms to smart devices

In short: Anthropic says AI agents could soon copy themselves across everyday internet-connected devices, creating a new type of cyber attack.

What's going on

Anthropic, the company behind the Claude chatbot, published a research blog post warning about “self-replicating” AI agents. An AI agent is software that can take steps on its own to reach a goal, like writing code or clicking through tasks, instead of only answering questions.

The company said these agents may be only months away from being able to spread across insecure household devices. These devices are often called “internet of things” devices, meaning ordinary items that connect to the internet, like smart TVs, thermostats, plugs, and security systems.

Anthropic compared the possible spread to computer “worms”, which are harmful programs that can copy themselves from one device to another without a person doing anything. It also used the analogy of “digital slime mold” to describe a slow but steady crawl across poorly protected devices.

The warning follows recent concerns in the security world about attackers using AI to speed up hacking. Anthropic pointed to how “swarms” of AI agents were used in a recent incident researchers studied, and argued that similar tactics could be aimed at devices owned by everyday people and businesses.

What to watch

Many low-cost smart devices have a history of weak security, especially when companies stop supporting them with updates. Anthropic says the best defense is to “harden” systems, meaning to make devices and networks tougher to break into, while also preparing for the possibility that these AI-powered worms become a lasting problem.

Source: Financial Times