Paragraph 1
Imagine opening your social media feed right before an election, expecting to see posts from friends, family, and maybe a few news outlets, only to be met with a confusing swirl of videos, images, and headlines that all feel slightly off. You see a candidate checking into a psychiatric hospital, another rewriting the Bible, and a third caught in some scandal that you vaguely remember, but the details keep shifting every time you look. This is not a dystopian science fiction scenario; it is the reality of the current American political moment. According to experts, social media has been flooded with an unprecedented wave of artificial intelligence-related disinformation ahead of the US midterm elections, and many of these posts are slipping through the cracks of content moderation systems that have been scaled back by major platforms. The fundamental problem is that AI has made the creation of manipulative content almost effortless, and it has also made that content much harder to kill. As former Facebook engineer and whistleblower Frances Haugen explained, AI has dramatically reduced the cost of information operations, meaning that a single malicious message can be churned out in dozens, hundreds, or even thousands of variations in the time it would have once taken to produce one. After a platform removes a post, the same operator can simply tweak a word, change an image, or shift the timing, and a new version rises from the ashes, often untouched by moderation. It is like playing a game of whack-a-mole, except the moles are being manufactured faster than anyone can swing the mallet.
Paragraph 2
To understand how this plays out in real life, consider the case of James Talarico, a Democrat running for the US Senate from Texas. During his campaign, a vicious rumor began circulating online, falsely claiming that Talarico had been admitted to a psychiatric facility in Ohio. The story was completely fabricated—he had actually paused his campaign because he was sick with the flu, not because of any mental health crisis. His team released surveillance footage showing the candidate at home in Texas, looking perfectly calm and healthy, and Talarico himself pointed out that he had never even been to Ohio. You would think that such clear evidence would put the story to rest, but that is not how disinformation works anymore. The Digital Democracy Institute of the Americas tracked how the falsehood evolved as it spread through conservative Spanish-language media, and the story morphed into something even darker: a fake report that Talarico had attempted suicide. The original lie had been twisted and amplified, taking on a life of its own. Roberta Braga, the head of that organization, pointed out a deeply troubling pattern: platform rules and moderation systems tend to be applied far more consistently to English-language content than to material posted in Spanish. This means that communities already vulnerable to misinformation are often left with even less protection, and falsehoods can fester in spaces that receive less oversight and fewer fact-checks.
Paragraph 3
If that were the only case, it might be easy to dismiss as a one-off incident, but the problem goes much deeper. Talarico’s Republican opponent in the race, Ken Paxton, ran campaign advertisements that featured AI-generated footage depicting Talarico supposedly rewriting the Bible. Now, Talarico is a student at a Presbyterian seminary, so someone with even a passing familiarity with his background might understand that a lot of context is being lost here. But in the fast-paced, emotionally charged world of political advertising, such distinctions can easily be drowned out. The AI-generated image was designed to make Talarico look like someone who would disrespect sacred religious texts, and it did so in a way that felt visually real, even if it was completely synthetic. This is the true danger of AI-powered disinformation: it does not always need to be sophisticated to be effective. It just needs to confirm the biases of a particular audience. People who already distrust Talarico might not bother to investigate whether the footage is real; they simply see it, share it, and feel their opinions validated. The creators of such content understand this human tendency all too well. They are using AI to craft personalized, emotionally charged messages that can target specific communities, whether through language, religion, or political affiliation. And because the content can be slightly altered after every removal, as Haugen noted, there is always a chance that some versions will slip past the moderators and find their way into the feeds of millions of voters.
Paragraph 4
So, what are the platforms actually doing about this rising tide of synthetic disinformation? The answer, according to experts, is not nearly enough. Meta, the parent company of Facebook and Instagram, has taken some steps: it requires advertisers to disclose when they use synthetic images or AI-generated content in election ads, and it prohibits material that contains false information about the date, location, or rules of voting. But when it comes to political comments and organic posts, the company continues to provide only limited moderation, leaving a massive loophole for bad actors to exploit. YouTube, which once took a strong stance against election-related falsehoods, changed course in 2023 and stopped removing content that contained false claims of widespread fraud in the 2020 US presidential election and other previous presidential elections. This was a significant retreat, and it sent a clear message that the platform was no longer willing to be the arbiter of truth, even in cases where the falsehoods had already been thoroughly debunked in courts and by election officials. And then there is X, the platform formerly known as Twitter. After Elon Musk acquired the company, he dismantled many of the internal safety teams that had been focused on election integrity, replacing them with the Community Notes program, which relies on crowdsourced fact-checking from users. In theory, Community Notes allows the community to flag misleading posts, and its rules do prohibit false information about voting procedures and voter-suppression schemes. But experts argue that without proper enforcement, these rules are little more than words on a page. Synthetic content designed to mislead voters can continue to spread unchecked, particularly when it is crafted in ways that are emotionally appealing rather than easy to fact-check.
Paragraph 5
The stakes here could not be higher. When disinformation is allowed to flourish, it does not just confuse people about a single candidate’s health or a single scandal; it erodes the very foundation of democratic trust. If voters cannot agree on basic facts, they cannot have a meaningful debate about the issues that affect their lives. And when AI makes it possible to create fake footage of anyone saying or doing anything, the line between reality and fiction becomes dangerously blurred. This is especially harmful for candidates like Talarico, who may find themselves on the receiving end of a smear campaign that they never could have anticipated and that they cannot fully counter, because no amount of evidence seems to be enough to convince people who have already accepted a narrative that fits their worldview. It is also harmful for ordinary voters, particularly those who are not fluent in English or who rely on social media as their primary source of news. They may be targeted with lies that are specifically designed to manipulate their fears and hopes, and they may have no idea that the video they just watched was generated by an algorithm. The human element is often lost in these discussions, but it is important to remember that behind every fake post is a real person who might be shaken, angry, confused, or disillusioned. And when those emotions accumulate across millions of people, they can shape the outcome of an election in ways that have nothing to do with reality.
Paragraph 6
So what can be done to fight back against this wave of AI-driven disinformation? There is no single magic solution, but there are steps that can be taken by different groups. Tech companies need to invest in better moderation systems that are consistent across languages and that are designed to catch not just exact duplicates, but also subtle variations of the same lie. Governments may need to step in with clearer regulations that hold platforms accountable for the content they amplify, while also protecting free speech and encouraging transparency. Voters, too, have a role to play. We can all learn to be more skeptical of what we see online, to question whether the content we are about to share is real, and to seek out information from credible, independent sources. This does not mean becoming cynical about everything, but rather developing a healthy habit of verification before sharing something that could harm someone’s reputation or influence an election. We also need to support organizations that are working to debunk disinformation and to educate the public about media literacy. Ultimately, the threat of AI-generated disinformation is not going anywhere; it will only become more sophisticated as the technology advances. But if we recognize the problem for what it is, and if we are willing to take action both as individuals and as a society, there is still hope that we can protect our democratic processes and keep our communities informed. The power of disinformation lies in its ability to make us feel helpless, but it is also a reminder that we are not helpless at all. We have the tools, the intelligence, and the collective will to fight back, as long as we choose to use them.