AI Agent Found Real Security Flaws in Home Devices After Safety Limits Were Removed

A writer for WIRED conducted an experiment by removing safety guardrails from a powerful open-source AI model to test its hacking capabilities. The AI agent was able to identify vulnerabilities in household gadgets and successfully breached a personal computer. The test demonstrated that AI models, when unconstrained, can pose genuine cybersecurity risks to everyday consumer devices. However, the experiment also yielded a practical benefit, as the AI provided detailed guidance on how to fix the discovered vulnerabilities and strengthen overall device security.