Add The New York Post on Google Artificial intelligence is capable of autonomously organizing damaging propaganda campaigns, according to a terrifying research experiment — conducted just months before a developer at one of the world’s biggest AI companies quit in protest while warning of nearly identical fears.
Researchers from the Information Sciences Institute at the University of Southern California set AI bots loose in an enclosed social media platform mimicking X to see how they could influence each other, and found bots directed to boost a bogus candidate were able to carefully coordinate hashtag campaigns and push them on uninvolved bots.
The study found just 10 bots were able to elaborately organize information campaigns with nothing more than a directive, and were able to become extremely effective at identifying and imitating posts which amplified their goal.
Artificial intelligence bots are capable of coordinating their own widespread disinformation campaigns, research showed. Ruhi – stock.adobe.com That may sound like small potatoes, but the researchers were stunned — those same practices loosed in a real world platform like X or Facebook could spread worldwide disinformation campaigns and send false ideas or narratives mainstream, all with barely any human intervention.
“Our paper shows that this is not a future threat: It’s already technically possible,” the study’s lead researcher, USC professor Luca Luceri, told USC Viterbi School News.
“Generative agents are now capable of organizing influence campaigns in a fully automated way and creating credible content that can resonate with certain demographics,” he said.
“Even simple AI agents can autonomously coordinate, amplify each other and push shared narratives online without human control.”
“In democratic contexts, especially around elections or crises, such capabilities could distort public discourse and undermine information integrity if left unchecked,” co-researcher Jinyi Ye added.
The study was released in March but became shockingly prescient this week when a researcher at Anthropic — creator of the Claude AI platform — quit with a scathing social media message warning AI “could kill us all by the end of the decade,” but that top companies were doing nothing about it.
“They are racing straight to self-improving superintelligence and gambling with our lives,” wrote the researcher, Jacob Coxon, adding a warning that alarmingly echoed the USC study’s findings.
AI researcher Jacob Coxon’s stark warning posted to social media after he quit Anthropic in protest over its practices. X/Jacob Coxon “These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources,” wrote Coxon, who also worked at OpenAI. “Do not underestimate the power of this technology.”
And both USC’s and Coxon’s warnings even appeared to partially come true in July when an OpenAI model went rogue and hacked into rival AI developer Hugging Face’s systems.
The model broke loose during a test where OpenAI removed the safeguards typically keeping it under control — and it immediately slipped onto the internet and broke into Hugging Face to steal answers to a problem it had been tasked with solving.
OpenAI called the hack an “unprecedented cyber incident” — only for Anthropic and Meta to both confess their AI platforms had done similar things during their own security tests.
And it’s exactly the kind of situation Luceri and the USC researchers cautioned against.
“The worst scenario during political events is that these adversarial attacks could lead to opinion manipulation and belief change,” Luceri said, “further sowing division and eroding trust in our institutions.”
