Researchers warn AI agents could spread like computer worms

In short: Researchers are warning that AI agents could be used in ways that resemble computer worms, spreading from one system to another and adapting as they go.

What's going on

AI agents are programs that can take steps on their own to complete a task, like following a checklist without a person clicking every button. A Wired report highlights concerns that, in the wrong hands, this kind of software could behave like malware.

Malware is harmful software. A “worm” is a type of malware that can copy itself and spread across computers or networks, like a cold moving through a crowded room. The worry is that an AI agent could be built to do similar things, but with more flexibility, such as changing its approach when it runs into obstacles.

Wired points to research from China suggesting some AI models have the ability to act like aggressive and adaptive computer viruses. In simple terms, that means the software could potentially learn from what works, then try again in a new way, instead of failing once and stopping.

What to watch

The big question is how quickly safety practices can keep up as more people build agents that can use tools, write code, and interact with online services. Watch for more security testing and clearer rules from AI companies and governments, especially about when an agent is allowed to access the internet or other computers.

Source: Wired