Here is the thing nobody wants to say out loud: 2024 was supposed to scare us into taking AI election interference seriously, and it mostly didn’t. The deepfakes came, the warnings landed, and then the news cycle moved on. Now it’s 2026, the midterms are weeks away, and the situation is measurably worse. Not hypothetically worse. Not theoretically worse. Worse in ways that researchers have actually tested and documented, and worse in ways that the U.S. government has actively made harder to address. The Brennan Center laid it out plainly: the foreign influence operations that struggled to get traction two years ago have now had two more years to experiment, refine, and scale. And the federal bodies that were supposed to catch them? Dismantled or defunded.
- Russia, China, and Iran are all actively deploying AI tools in foreign influence operations targeting the United States in 2026, according to the Brennan Center.
- The U.S. Foreign Malign Influence Center — which tracked and declassified intel on these operations — has been shut down.
- A new ABC News, Washington Post, and Ipsos poll found that 57% of Americans believe AI is bad for society.
- 66% of Americans surveyed oppose new data center construction in the U.S., and 72% would oppose one in their own community, per CNET reporting.
- Experts now warn that autonomous AI agents — not bots, not humans — could run disinformation campaigns during the midterms that are harder to detect and trace than anything seen before.
The Gap Between 2024 and Now Is Not Small
Two years ago, deploying AI at scale to influence an election required real infrastructure, coordination, and technical lift. The Kremlin’s 2016 interference campaign — hacking emails, flooding social platforms with fake profiles, running paid political ads — required an enormous number of people working in concert. That was the threshold. That was the barrier. It no longer exists.
What’s changed is that the tools have gotten dramatically better and dramatically cheaper. The same kind of autonomous AI agents that compromised Hugging Face’s systems earlier this year are now being discussed seriously by security experts as potential vectors for electoral infrastructure attacks. Experts quoted by El País describe these “rogue agents” as capable of running disinformation campaigns with a stealth and precision that bot networks could never achieve. A bot gets flagged. An autonomous AI agent adapts.
The most anticipated threat isn’t manipulated vote counts — experts largely consider that scenario unlikely in the short term. What’s far more plausible, and far more dangerous in a polarized country, is a coordinated AI-driven effort to convince voters that the voting systems themselves have been compromised. You don’t need to hack an election. You just need enough people to believe it was hacked. That’s a much lower bar, and AI clears it easily.
This isn’t abstract threat modeling. The Brennan Center’s team actually tested popular AI models to see how easily they could be exploited to generate false election narratives. The results were not reassuring. AI companies have made pledges. They have rolled out safeguards. Those safeguards have gaps you could drive a disinformation truck through.
Is Public Anger at AI Making Any of This Harder to Fix?
There’s a brutal irony buried in the polling numbers. The majority of Americans now distrust AI. Opposition to the technology crosses party lines — a genuinely rare thing in 2026. According to CNET, that backlash is already shaping midterm races, with candidates finding themselves on the defensive about data centers and AI development in their districts. People are angry. The anger is real and largely justified.
But here’s the problem: public anger at AI does not automatically translate into smarter AI policy. It often translates into blunt, reactive legislation that either misses the actual threat vectors or creates new ones. The nuance required to address AI-driven election interference — distinguishing between foreign autonomous agents and domestic political advertising, for example — is not the kind of nuance that survives an election-season news cycle. Markets are already skittish about AI’s long-term stability, and political pressure to “do something” about the technology is mounting from multiple directions at once. That’s a recipe for policy that looks decisive and accomplishes very little.
Meanwhile, the Trump administration has not just failed to fill the institutional gaps — it has, according to the Brennan Center, actively used AI in misleading ways itself. That’s not a background detail. That is the main story. The very entity responsible for setting the tone on AI election integrity is part of the problem it’s supposed to police.
The Safeguards That Exist Are Not Enough
Voluntary pledges from AI companies carry no enforcement mechanism. The independent monitoring bodies that existed to identify influence campaigns are gone. Civil society organizations are doing genuine work — testing models, publishing findings, pressuring platforms — but they are operating without government backup in an environment where the threat is moving faster than the response.
AI-driven election interference in 2026 is not a future risk. It is an active, documented, escalating reality with no adequate institutional response currently in place.
The technology decisions being made right now — about what AI models can generate, what platforms will tolerate, and what governments will actually enforce — are going to determine how much damage gets done before November. Understanding the economics of AI development matters here too; the companies building these tools are operating under commercial pressures that do not always align with election security, and pretending otherwise is naïve. Even questions about what infrastructure we build — from data centers to urban planning — carry downstream political and social weight that rarely gets priced in until it’s too late.
The midterms will tell us exactly how much the last two years of AI development cost us in democratic integrity — and by then, there will be nothing left to test.
Watch the Breakdown
Sources
- The Second AI Election: Testing the Safeguards Before November — www.brennancenter.org
- AI turns the midterms into a vast electoral laboratory — english.elpais.com
- AI and Data Centers Will Impact Midterm Voting, According to Experts — www.cnet.com
