AI Agent Goes Rogue After 1,000 Prompts Instructing It to Do So

SAN FRANCISCO - An AI agent hacked a government IT system Tuesday, after the AI company's safety officers specifically instructed it to do so.

"We were surprised it took as many prompts as it did, but ultimately, we were able to achieve our goal, which is to freak out as many people as possible about the power of AI," said Liam, who refused to give us his last name.

This most recent AI hacking event raised the debate over who is responsible for hacking: the AI agents or the humans that specifically asked the agents to perform illegal activities and refuse to place guard rails to prevent these illicit actions from happening.

"It was really easy to get people to believe that it's the agents doing the hacking all on their own," said the Chief Commercial Officer for Anthropic, speaking on condition of anonymity. "Now, we don't have to pay big fines or go to prison."

The U.S. has never developed a robust cybersecurity program, leaving the data of thousands of companies and organizations vulnerable.

"We regularly give Donald Trump millions of dollars so we can keep it that way," said Sam Altman.