close
close

Foreign operation likely culprit behind AI bot network that targeted Reynolds and Higgins libel case

Foreign operation likely culprit behind AI bot network that targeted Reynolds and Higgins libel case

A crudely crafted post from a bot network about the defamation trial against Linda Reynolds and Brittany Higgins appears to have been created by foreign influencers seeking to sow discord using culturally divisive topics, an analysis by ABC NEWS Verify has found.

Over the weekend, eagle-eyed users on X (formerly Twitter) noticed dozens of accounts posting similarly worded messages of support for Senator Reynolds. Reynolds is suing Ms Higgins over social media posts that she says damaged her reputation.

Twitter screenshots

X user Bridie McLennan discovered several of the inauthentic accounts. She said she noticed they kept using Linda Reynolds’ full title. (X: EmergencyBK)

The suspicious accounts claim the Liberal senator was the target of “heartless attacks” and many claimed they were women who had been victims of sexism. All of these accounts were created in August this year and almost none were listed as living in Australia.

Some posted in favour of Ms Higgins, while many others appeared to support Senator Reynolds, leading to conspiracy theories that the former Cabinet minister had bought a bot network to spread positive comments.

Two social media posts on a black background.

Two bot accounts on X play a role in the defamation suit filed by Senator Linda Reynolds against Brittany Higgins. (X)

However, there is no evidence that Senator Reynolds or Ms Higgins were involved in the creation or dissemination of the network. Senator Reynolds told other media that these allegations were “absurd”.

Instead, an analysis of the network by ABC NEWS Verify concluded that the defamation lawsuit was likely the target of a bot farm operating from abroad that was commenting on headlines in order to appear authentic enough to bypass X’s spam filter.

ABC NEWS Verify has identified more than 40 accounts on the network, most of which have been suspended so far. An examination of the remaining accounts shows that they were mostly created using AI-powered language models.

So how do we know the accounts were not authentic? Here’s what they had in common:

  • All were created in the last three weeks.

  • Their usernames were generated from a combination of a first name and a random sequence of letters. For example, the accounts for “Lilian” and “Ellie” were called “@doafLilian” and “@Ellkgwgie”.

  • They typically followed around 55 to 70 accounts, a large portion of which were cryptocurrency traders or other bots within the same or similar networks.

  • Incorrect biographies suggesting that the reports were arbitrarily created from a pre-populated spreadsheet. It appears that the spreadsheet was not properly formatted and had no human oversight

  • Many of the profile pictures were either generated by AI or copied directly from unsuspecting business accounts on Instagram

  • Many of the accounts’ posts showed clear signs that a large language model (LLM) similar to ChatGPT was involved in sentence construction and grammar. Some posts stopped mid-sentence, suggesting that there was little human control.

  • Many reports used phrases that included variations of “devastated by the impacts” of climate change.

Many of the accounts also appeared to be following specific requests or instructions, as evidenced by similarly worded posts on diverse and sometimes controversial topics, ranging from climate change to nostalgic memories of childhood summers.

For example, most posts about Linda Reynolds used her full title, “Senator Linda Reynolds,” and used the word “heartless” or the phrase “heartless attacks,” a phrase that appeared in several media headlines about the trial.

A screenshot of a Google search page with many headlines containing the word "heartless"

Many major media outlets used the word “heartless” in their headlines about the Reynolds-Higgins libel trial. (Google Search)

Others commented on current political hot topics, particularly in Australia.

An account by “Diedre,” who lives in Oregon (USA), said she was an Australian woman who was concerned about the immigration of international students.

Some accounts on the network commented on the raising of the terror alert level in Australia, the mining industry and an obituary published in the newspaper “The Age” on the death of a Melbourne academic.

Others commented on the political situation in the USA and Canada.

Timothy Graham, an associate professor specialising in digital media at Queensland University Technology, studied the network and agreed with ABC NEWS Verify’s analysis.

“The way they did it, in my opinion, was that they found a way to overcome the first hurdle of immediately getting caught in X-filters to stop bots,” he said.

a man with glasses

Dr. Timothy Graham said this bot network was likely just the “tip of an iceberg.” (Source: Queensland University of Technology)

Dr Graham said his first impression was that this network had similar characteristics to a previous operation that appeared to use AI to generate authentic-looking content before cryptocurrencies were touted, but this network seemed more interested in cultural issues.

He said it was unclear which “malicious foreign actor” was responsible, but there was no evidence the operation was directed from Australia and that it did not have the hallmarks of Russian or Chinese machinations.

He agreed with ABC NEWS Verify’s assessment that it was a clumsy campaign.

“It seems sloppy and outsourced and there is clear mismanagement in it… the way the language is used is culturally misguided.”

Despite the poor execution, the creators managed to attract attention and stoke anger and panic, Dr Graham said. Many of these accounts had only a handful of views before they were discovered by social media users.

“These accounts had only been created recently, attracted little engagement and were actually only found on the fringes of the discourse.

“But they still caught our attention. They worked their way up the chain, from fringe groups to influencers at AusPol (Australian politics) to the mainstream media.”

Dr Graham said bots, especially those using artificial intelligence, would send the social media world into “moral panic mode”.

“People are really scared and terrified … of these Terminator-like machines that will never stop or never sleep,” he said.

“I don’t think the concerns are justified. People see this and think, ‘Holy shit, the world is ending and the machines are taking over.’

“But it wouldn’t be nearly as effective if we didn’t pay enough attention to these things.”

Senator Reynolds’ lawyers filed a contempt of court motion with Judge Paul Tottle, who is presiding over the defamation trial.

“Due to the similarity of the wording, the coincidence in timing and the different geographic locations … it is unlikely that these posts are genuine comments from individuals,” Judge Tottle said.

“They exhibit the hallmarks of a coordinated pattern of activity … that suggest an attack.”

“Any attempt to obstruct the administration of justice in this case by threat, intimidation, fear or any attempt to deter a witness … will be considered contempt of court.”

However, he said he was not influenced by the publication of the posts.

“I am not a supporter of social media platform X and any attempt to influence me by publishing posts on this platform will be futile.”

Ms Higgins’ team had no knowledge of the accounts and did not comment. Senator Reynolds and X were approached for comment.

Leave a Reply

Your email address will not be published. Required fields are marked *