Washington — When the U.S. announced the seizure of 32 internet domains tied to Russian efforts to ply American voters with disinformation ahead of November’s presidential election, prosecutors were quick to note the use of artificial intelligence, or AI.
The Russian operation, known as Doppelganger, drove internet and social media users to the fake news using a variety of methods, the charging documents said, including advertisements that were “in some cases created using artificial intelligence.”
AI tools were also used to “generate content, including images and videos, for use in negative advertisements about U.S. politicians,” the indictment added.
And Russia is far from alone in turning to AI in the hopes of swaying U.S. voters.
“The primary actors we’ve seen for election use of this are Iran and Russia, although as various private companies have noticed, China also has used artificial intelligence for spreading divisive narratives in the United States,” according to a senior intelligence official, who spoke on the condition of anonymity in order to discuss sensitive information.
“What we’ve seen is artificial intelligence is used by foreign actors to make their content more quickly and convincingly tailor their synthetic content in both audio and video forms,” the official added.
But other U.S. officials say the use of AI to spread misinformation and disinformation in the lead-up to the U.S. election has so far failed to live up to some of the more dire warnings about how deepfakes and other AI-generated material could shake-up the American political landscape.
“Generative AI is not going to fundamentally introduce new threats to this election cycle,” according to Cait Conley, senior adviser to the director of the Cybersecurity and Infrastructure Security Agency, the U.S. agency charged with overseeing election security.
“What we’re seeing is consistent with what we expected to see,” Conley told VOA.
AI “is exacerbating existing threats, in both the cyber domain and the foreign malign influence operation-disinformation campaigns,” she said. But little of what has been put out to this point has shocked officials at CISA or the myriad state and local governments who run elections across the country.
“This threat vector is not new to them,” Conley said. “And they have taken the measures to ensure they’re prepared to respond effectively.”
As an example, Conley pointed to the rash of robocalls that targeted New Hampshire citizens ahead of the state’s first in the nation primary in January, using fake audio of U.S. President Joe Biden to tell people to stay home and “save your vote.”
New Hampshire’s attorney general quickly went public, calling the robocalls an apparent attempt to suppress votes and telling voters the incident was under investigation.
This past May, prosecutors indicted a Louisiana political consultant in connection with the scheme.
More recently, the alleged use of AI prompted a celebrity endorsement in the U.S. presidential race by pop star Taylor Swift.
“Recently I was made aware that AI of ‘me’ falsely endorsing Donald Trump’s presidential run was posted to his site,” Swift wrote in an Instagram social media post late Tuesday.
“It brought me to the conclusion that I need to be very transparent about my actual plans for this election as a voter,” she wrote, adding, “I will be casting my vote for Kamala Harris and Tim Walz.”
But experts and analysts say for all the attention AI is getting, the use of such technology in attacks and other influence operations has been limited.
“There’s not a tremendous amount of it in the wild that’s particularly successful right now, at least to my knowledge,” said Katie Gray, a senior partner at In-Q-Tel, the CIA’s technology-focused, not-for-profit strategic investment firm.
“Most attackers are not using the most sophisticated methods to penetrate systems,” she said on September 4 at a cybersecurity summit in Washington.
Others suggest that at least for the moment, the fears surrounding AI have outpaced its usefulness by malicious actors.
‘We jump to the doomsday science fiction,” said Clint Watts, a former FBI special agent and counterterror consultant who heads up the Microsoft Threat Analysis Center (MTAC).
“But instead, what we’re seeing is the number one challenge to all of this right now is access, just getting to the [AI] tools and accessing them,” he said, speaking like Gray at the cybersecurity summit.
Over the past 14 months, MTAC has logged hundreds of instances of AI use by China, Russia and Iran, Watts said. And analysts found that Moscow and Tehran, in particular, have struggled to get access to a fully AI toolbox.
The Russians “need to use their own tools from the start, rather than Western tools, because they’re afraid they’ll get knocked off those systems,” Watts said.
Iran is even further behind.
“They’ve tried different tools,” Watts said. “They just can’t get access to most of them for the most part.”
U.S. adversaries also appear to be having difficulties with the underlying requirements to make AI effective.
“To do scaled AI operations is not cheap,” Watts said. “Some of the infrastructure and the resources of it [AI], the models, the data it needs to be trained [on] – very challenging at the moment.”
And Watts said until the products generated by AI get better, attempted deepfakes will likely have trouble resonating with the targeted audiences.
“Audiences have been remarkably brilliant about detecting deepfakes in crowds. The more you watch somebody, the more you realize a fake isn’t quite right,” according to Watts. “The Russian actors that we’ve seen, all of them have tried deepfakes and they’ve moved back to bread and butter, small video manipulations.”
…