Beijing-based ‘Green Cicada’ AI network uncovered on social media, fears of US election disruption
In short:
A social media bot network linked to a Chinese university and AI company has been uncovered on X.
The information warfare network, dubbed “Green Cicada”, is one of the largest identified.
What’s next?
There are concerns Green Cicada has been staged to disrupt the coming US presidential election.
A network of at least 5,000 AI-run accounts has been exposed in a suspected Chinese-run information warfare campaign to spread divisive political discourse on the social media platform X.
Local cyber security company CyberCX says it has uncovered an operation linked to a Chinese university and AI company that appears to mainly target contentious American narratives but has sometimes also engaged with Australian content.
“While the information operation capability is currently relatively ineffective, we assess it could be leveraged to conduct harmful activities in future,” the company warns in a newly completed report.
Researchers believe the cluster of at least 5,000 unauthentic X accounts, dubbed the Green Cicada Network, is almost certainly controlled and coordinated by an artificial intelligence Large Language Model (LLM)-based system.
An employee of a Beijing-based AI company who studied at Tsinghua University, which has close links to the People’s Liberation Army and Beijing’s intelligence apparatus, has been identified as the person likely to have established the emerging operation.
“The network is increasingly engaging in political discourse, but most accounts remain dormant,” the report states in findings that have already been shared with various federal government agencies.
‘Green Cicada’ accounts could be staged to disrupt US election
While the Green Cicada Network predominantly engages with US political and cultural issues, it has also been observed amplifying hot-button political issues in Australia, the UK, Western Europe, India, Japan and other democratic countries.
“We observed limited amplification of Australia-specific issues and posting from purportedly Australian personas. Amplified issues include support or opposition to political candidates, nuclear energy, economics, housing, migration, protests and foreign policy,” CyberCX found.
“Here we have a fake network that is infiltrating our democratic discourse and trying not necessarily to support one side or the other of these debates, but trying to drive a wedge [between] sides of this debate, trying to deepen division and deepen polarisation,” spokesperson Katherine Mansted said.
According to CyberCX, the network “may plausibly be staged to interfere in the upcoming presidential election,” with the company saying it has observed it “improving operational execution over time and sharply increase activity since July 2024”.
The cluster of up to 8,000 unauthentic accounts is considered one of the largest publicly exposed to date and may be the first significant China-related information operation to use generative AI at the core of its activities.
“We assess that if the full scale of available accounts were engaged, they could successfully amplify polarising content to sow division and undermine trust in civil institutions,” the report concludes.
A number of the accounts were identified by forcing malfunctions using prompt injections, which override an AI model’s original instructions.
Analysis of a cluster of Green Cicada accounts found the account they most frequently engaged with was that of X’s owner, Elon Musk.
Last month, Australia and key regional partners accused a Chinese spy agency of cyber espionage, targeting government and business networks, in a large-scale operation that involved stealing hundreds of usernames and passwords.
