I Let an AI Agent Hack All My Gadgets—and I’d Do It Again
wired ·
A journalist decided to test an unguarded, 'abliterated' AI model by letting it attack his home network. The agent quickly found vulnerabilities in his printer, Wiim stereo, and other IoT devices, and it even hacked into a PC using a cryptographic key after other login attempts failed. It also flagged security problems in his 'vibe-coded' projects, such as exposed API credentials. Instead of causing harm, the agent offered practical advice for improving security, like updating firmware and using a guest network for smart devices. The journalist reflects that while using such a rogue agent feels scary, having an AI hacker of your own may be the best way to defend against AI-driven attacks. Experts quoted in the article suggest that widespread availability of cyber-capable AI could ultimately make software more secure, though critical systems may need stronger protections.
wired