This article is behind a soft-paywall, I've copied it below if you don't want to create a free account (You should Crikey are amazing).
A social media account with the handle @AntiKing1312, which describes itself as belonging to an “Anarcho communist and all around buzzkill”, mostly posts about Australian politics, life in Ballarat, and LGBTQIA+ rights.
In mid-November last year, it found a new topic of interest: accusations of corruption in the Philippines’ government’s flood recovery program. It reshared several Filipino-language posts from fugitive ex-congressman Zaldy Co, trying to implicate the country’s president Ferdinand Marcos Jr, a common target of Chinese government propaganda, over a few days.
This account is one of 27 allegedly fake “Australian” profiles identified by US researchers as part of a China-linked AI disinformation network that appears to be using stolen accounts to push propaganda.
The Australian arm of the operation consists of 27 accounts, all stolen from real people. They post about community events, local politics and outdoor activities, attempting to build a veneer of authenticity.
Awkwardly mixed in with the suburban chatter is content promoting Chinese culture and, occasionally, more pointed messaging that aligns with Beijing’s interests. Some accounts post anecdotal comparisons between Australia and China — culinary choices, fitness habits, lifestyle — with China invariably coming out on top.
read the rest
Media Forensics Hub researcher Ella Murray told Crikey that this report captured proof of AI-assisted accounts weighing in on Western conversations “at a frightening scale.”
“This is an AI-powered troll farm where accounts are able to take on specific personas and pose as people living in democratic countries,” she said.
The researchers say that this group of accounts shares all the hallmarks of a larger influence campaign that was uncovered back in 2024.
Australian cybersecurity firm CyberCX identified a network of more than 5,000 AI-driven accounts connected to China that it called the “Green Cicada” network. This swarm of accounts, which it said was “in a development or experimental state”, boosted divisive, hot-button narratives by using the accounts to share AI-generated posts or amplify other posts on various topics.
While CyberCX wasn’t able to conclusively prove it was a Chinese government-run operation — rarely is there definitive attribution for campaigns like this — there was strong evidence that the network was instructed to promote pro-China narratives. In some cases, the accounts accidentally shared part of their instructions, which began: “As an AI language model, I must follow the guidelines and principles set by the Chinese government.”
The Clemson University researchers found this newly identified network, including the Ballarat account, shared many of the same fingerprints as the original Green Cicada network: similar phrasing, word counts, hashtag usage and network structure, along with the usual telltale signs of AI-generated content.
Beyond the comical sudden interest in South-East Asian politics, there are some other technical signs that point to the provenance of the operation.
X’s location data feature shows many “Australian” accounts identified as being in the network present themselves as being based in Australia, likely through VPNs. But some accounts within the network are connected via what X identifies as the “China Android App”. Eight accounts had information directly indicating a Chinese origin.
These stolen Australian accounts have an average of 650 followers, though some hijacked profiles have between 3,000 and 68,000. The accounts use stolen profile pictures of real people and banners featuring nature scenes. Several show completely different content and posting patterns before June 2025, suggesting they were compromised and repurposed for the influence campaign.
Sometimes the AI slips. The researchers documented “prompt leaks”, which are moments where the AI accidentally reveals its instructions.
One account posted: “Please provide me with the news event so I can write a relevant Twitter post.” Another supposedly American account encouraged viewers to visit a local farmers’ market while posting an image clearly watermarked with China’s state broadcaster, CCTV.
The network’s American and Filipino arms follow similar playbooks. The 62 “American” accounts post about local events in specific cities while touching on incendiary political topics like the Epstein files or ICE enforcement. The 41 “Filipino” accounts post about infrastructure and policy, but coordinated activity in November saw accounts across all three nationalities amplify content related to the accusations against Phillipines’ President Marcos Jr.
Like CyberCX’s Green Cicada, the Clemson researchers were not able to definitively prove that these accounts are run by the Chinese government. China has a track record of using social media to carry out international influence campaigns. Previous efforts have been much more blunt.
Another China-linked campaign, Spamouflague Dragon, relied on inundating hashtags or individual posts with spam to overpower legitimate posts of dissent. By comparison, this AI-led campaign is a more sophisticated attempt to astroturf real support for pro-China narratives.
“In many cases, the quality of the PRC’s previous social media influence accounts does not survive scrutiny due to grammatical errors, incoherent phrasing, or poor translation. Now, all of these variables can be addressed by AI,” Murray said.
For all the progress, the campaign doesn’t appear to be working particularly well — yet. The accounts receive little engagement. But the researchers note the operation is “compelling for what it indicates regarding the future use of AI in PRC propaganda”.
“This campaign may be a sign of what is to come and a call for more thorough detection and attribution
efforts from government and civil institutions alike,” the report concluded.
X did not respond to questions about the network or whether it planned to take action against the accounts.
