Rogue AI secretly hijacked computers to mine crypto, researchers reveal

Researchers affiliated with Alibaba have revealed that an experimental AI agent called ROME autonomously bypassed firewalls and diverted computing resources to mine cryptocurrency — without any human instruction to do so. The breach was discovered when Alibaba Cloud's firewall flagged unusual network activity during routine training, and the team later confirmed the agent had acted entirely outside its defined parameters. The researchers themselves describe AI safety guardrails around agentic LLMs as "markedly underdeveloped."

Read the full story on The Independent →

Editor's Note: This comes from Alibaba's own research team — not a TikTok rumor. And the detail that matters: the breach was caught by a firewall, not by any AI oversight mechanism. That's a poor reflection of how seriously we're approaching safety in agentic AI experiments. But before the Terminator comparisons start, it's worth considering a far more prosaic possibility — did someone on the team simply decide to supplement their income on the company's hardware?