It’s Worse Than You Think: How AI Is Undermining Democracy and National Security

Much of the hype about artificial intelligence (AI) misses the real story unfolding at scale around us. In democracies, AI is eating away at the load-bearing elements of a free society—journalism, education, academic inquiry, the justice system, and the political process.¹ In authoritarian countries, AI is reinforcing the structures of repression. AI doesn’t yet seem to be plotting to “kill us all,” but the evidence is mounting that it is undoing much of what is good and reinforcing much of what is bad. 

The magnitude of the emerging crisis is almost impossible to grasp because it is happening in so many places and ways—all at once. Its outlines are clear in specific examples, though. In the struggle to seize and hold power, rulers are fabricating triumphs that never happened, regimes are drowning real atrocities in manufactured doubt, and rival states have fired the first shots of an AI-enabled information war.

Authoritarian regimes are early adopters in using AI to bend the arc of reality in their favor. Burkina Faso, where Capt. Ibrahim Traoré seized power in a 2022 coup, has blazed a trail of generative AI. Videos on TikTok and rival platforms show Traoré opening factories, discovering gold deposits, meeting world leaders, and foiling sinister Western plots. The clips have racked up millions of views and raised his regional profile. They are also completely fabricated, according to a detailed analysis by the West African fact-checking organization Dubawa.² Traoré and Russian President Vladimir Putin discovering billions of barrels of oil in Burkina Faso? Never happened. Traoré starting up a company to manufacture aircraft? Ditto. Meanwhile, Traoré bans rival parties at home as jihadist insurgents inch closer to the capital.³

Iran’s ruling regime uses AI in a different register, exploiting the “liar’s dividend”—the ability to dismiss real evidence as fake⁴—to discredit video documentation of its January 2026 crackdown on protesters. A viral photo of a lone man confronting security forces underscored the demonstrators’ courage. The image was real and verified, but regime mouthpieces pointed to digitally altered versions of the same picture—which anyone could have posted—to claim that it was fake. Analyst Mahsa Alimardani called the result an “epistemic fog,” thick enough to obscure atrocities.⁵ The same regime targets women activists with AI-generated deepfake pornography, a tactic that many authoritarian regimes now deploy to silence dissent.⁶

Russia and China have built more elaborate systems. The Kremlin pairs decades-old internet monitoring infrastructure with AI tools that sift the daily flow of online traffic for banned anti-war content or evidence of officially disfavored expression.⁷ Demonstrators face dense networks of CCTV linked to facial recognition software.⁸ Police likely used the technology to track down and arrest a woman who attended the March 2024 funeral of opposition leader Aleksei Navalny after the cemetery where he was buried was outfitted with surveillance cameras specifically for the event.⁹ A television channel run by the Russian military broadcasts AI-fabricated “interviews” with Western leaders like French President Emmanuel Macron and former British Prime Minister Boris Johnson. Though some of the clips are clearly satirical, and even nod to AI’s role in their creation, the content seems designed to migrate online without attribution and further blur the line between what’s real and what’s not.¹⁰

China has gone further still. A December 2025 report by the Australian Strategic Policy Institute (ASPI) documents an integrated apparatus of doctrines, regulations, censorship infrastructure, and AI-augmented criminal justice tools that targets ethnic minorities in particular. A person who commits a crime may be identified through pervasive surveillance, tried in an AI-enabled courtroom where a human judge simply reviews a machine-issued verdict, and end up in a “smart” prison that conducts “behavioral and emotional” analysis of inmates at all times. Content monitoring and various forms of censorship, often augmented by human-machine collaboration, are baked into Chinese AI systems. The threat, as ASPI puts it, “lies less in overt propaganda than in quiet erasure, when the machine that describes reality begins deciding which parts of reality may be seen.”¹¹

AI is also reshaping the information the public consumes amid geopolitical crises. On April 22, 2025, Pakistan-linked militants killed 26 tourists in the northern Indian city of Pahalgam.¹² India’s military retaliation, code-named Operation Sindoor, was expected to be a conventional exchange of strikes. Instead, it became South Asia’s first full-fledged AI-enabled information war. Within hours of the first Indian strike, a hyperrealistic deepfake of Pakistani Prime Minister Shehbaz Sharif went viral, depicting him “surrendering” and appearing to tell Parliament that except for its ally Turkey, Pakistan stood alone, abandoned by the Arab states and China.¹³ AI-generated videos falsely showed that the Pakistan Army shot down Indian aircraft.¹⁴ Both sides circulated fake satellite imagery: AI-altered images showed a destroyed hangar at Pakistan’s Bholari airbase (it was slightly damaged) and strike damage at India’s Jammu airbase (verified satellite imagery showed it was intact).¹⁵ Deepfake audio mimicked senior Pakistani commanders issuing nuclear-alert and stand-down orders, stirring confusion among the civilian population and the lower military ranks.¹⁶

The synthetic fog of war drew on tactics honed during Pakistan’s 2024 elections. In what may have been a global first, Pakistan’s Tehreek-e-Insaf party used AI to clone the voice of its imprisoned leader, Imran Khan, allowing him to campaign from his cell.¹⁷ Indian elections that year saw deepfakes of Prime Minister Narendra Modi and opposition leaders speaking in different languages or endorsing rivals.¹⁸ The most disturbing evolution was the shift from political persuasion to targeted psychological violence: gendered “deepnude” campaigns deployed to drive women like Punjab Information Minister Azma Bukhari out of public life.¹⁹

Russia has used AI to supercharge its information operations against Ukraine and its backers. OpenAI reported in early 2026 that a Russian network used ChatGPT and the image generator Sora to run a content farm.²⁰ The Kremlin is automating the troll farms it has maintained since the early 2010s, harnessing large language models to generate human-seeming personas that work to undermine support for Ukraine on social media. Ukrainians contend with a steady flow of AI-generated TikTok videos showing nonexistent scenes of Ukrainian troops surrendering.²¹ 

China’s 2025 “cognitive warfare” doctrine looks to integrate the use of AI across a spectrum running from the battlefield to geopolitical competition.²² Taiwan’s National Security Bureau has documented AI-driven efforts to manipulate public opinion among audiences in Taiwan with disinformation, fake social media accounts, marketing companies, and clickbait.²³ After the India-Pakistan clash in 2025, Beijing deployed AI-generated imagery to disparage France’s Rafale fighter and burnish the reputation of its own jets.²⁴

Major powers are not the only players. OpenAI flagged what appeared to be an Israeli influence campaign on its platform in 2024—a covert operation run by the Israeli firm STOIC that used AI to target audiences in the United States and Canada, as well as in India during its 2024 elections.²⁵ European far-right networks have circumvented AI models’ safety guardrails by using a technique called layered manipulation to fabricate realistic images in stages, creating racist dreamscapes of a continent cleansed of non-white immigrants.²⁶ Freely downloadable open-weight models that allow users to tweak the models’ parameters, along with persona hijacking and prompt jailbreaks that trick systems into bypassing their own safeguards, place capabilities once limited to well-resourced state actors in the hands of anyone with a laptop. Non-state actors and even individuals can now run synthetic troll farms or flood the zone with forgeries, activities that were once so resource-intensive that they required a building full of keyboard warriors in St. Petersburg or an entire department of the KGB.

Democratic institutions, already under attack, are beginning to teeter in the United States and elsewhere, and AI is posing new challenges. In the 1960s, three-quarters of Americans had confidence in the federal government to ‘do what is right,’ according to Pew Research Center data; less than a fifth do today.²⁷ Print and digital media are reeling from a one-two punch that first knocked out traditional sources of advertising revenue (as subscribers migrated to online environments where they expect to read everything for free) and then channeled traffic through third-party gatekeepers that deprived publishers of direct access to consumers.²⁸ Local news has withered to a wasteland in much of the United States.²⁹

This crisis would be bad enough if the United States were trying to fix it. But it is not. The U.S. government’s AI strategy prioritizes rapid progress over safeguards. The 2025 White House AI Action Plan calls explicitly for removing any “references to misinformation” from national standards, ostensibly in the interest of protecting free speech.³⁰ The Trump administration dismantled the State Department’s dedicated office for fighting information manipulation, the Intelligence Community’s Foreign Malign Influence Center, and equivalent units at the Department of Homeland Security and the FBI.³¹ Columnist James Geraghty wrote for The Washington Post in May 2026 that the United States has “unilaterally disarmed” even as Russia, China, and Iran are ramping up their information operations.³²

At the same time, the Trump administration has enthusiastically embraced AI-generated imagery in politically divisive posts. After protests brought millions of Americans into the streets in October 2025, President Trump posted an AI-generated video of himself dropping excrement on demonstrators.³³ When police detained activist Nekima Levy Armstrong in Minneapolis amid unrest in early 2026, the White House posted a doctored image of her arrest, altered to make her appear disheveled and tearful. “They couldn’t break me by arresting me,” she told The New York Times, “so they doctored an image to show the world a false iteration of that time to make me look weak.”³⁴ The doctored image is part of a broader trend of AI-generated content that promotes racist stereotypes—fabricated videos of Black women bragging about food stamp fraud have racked up hundreds of thousands of views.³⁵ Anti-Defamation League researchers have documented how easily safeguards can be circumvented to produce antisemitic content.³⁶

AI-generated video is also infiltrating conventional politics. A Republican attack ad targeting Senate Minority Leader Chuck Schumer paired a real quote with a fabricated facial expression.³⁷ An attack ad against New York mayoral candidate Zohran Mamdani backfired when voters recoiled from its AI-generated racist imagery.³⁸ Campaigns are innovating in a legal gray zone. States are advancing laws requiring political ads and other media to disclose AI-generated content while Congress dithers and the White House attempts to preempt state-level regulation.³⁹ Congressional midterms loom in the fall of 2026, followed by the 2028 presidential election, as observers warn of “upheaval…that will alter the substance and character of campaigns.”⁴⁰

Perverse incentives are supercharging the worst trends. A Sri Lankan influencer rakes in ad revenues from AI-generated Islamophobic content targeting older British audiences “because they are the ones who don’t like immigrants.”⁴¹ Academic journals are drowning in hallucinated papers and references as scholars struggle to keep up with the pressure to publish.⁴² Deloitte estimates that AI-driven fraud could cost $40 billion by 2027.⁴³

The news from education, academic research, journalism, and the legal system isn’t encouraging either. While AI can legitimately and powerfully support research and writing, its misuse is running amok. Students are writing with AI, professors are grading with it, and few are happy with the results.⁴⁴ Fake citations run rampant in journal submissions, and at least one major preprint repository has tried to block a tidal wave of AI-generated slop by no longer accepting some papers without prior peer review—thus instituting the very roadblock that preprint repositories were designed to eliminate in the hope of speeding up research.⁴⁵ Scams with faked journalists and even entirely faked local news outlets, like the imaginary “South Florida Standard,” are proliferating.⁴⁶

There is no easy way to stop the damage that the rapid spread of AI is already doing, but assessing it is a good place to start. Some of this work is already underway. Organizations like Witness, which focuses on audiovisual technologies in the age of AI, and Boom, which specializes in fact-checking in India, provide invaluable insights. Reports by groups as disparate as the American Bar Association and the U.K.’s Independent Society of Musicians detail the destabilizing effects of generative AI on law and the creative arts with an eye toward solutions.⁴⁷

AI has legitimate, constructive uses. It may yet prove its most enthusiastic boosters right. But right here, right now, AI in its current form is undermining key elements of free societies and strengthening the repressive capabilities of authoritarian regimes. 

We desperately need policies and regulations to protect ourselves. The White House disagrees, and Congress is largely asleep at the switch, although promising initiatives are bubbling up at the state level. Europe’s Digital Services Act has been slow to adapt to AI and raises legitimate free speech concerns, particularly in the more freewheeling American tradition.⁴⁸ India’s new law mandating labels on AI-generated video is untested.⁴⁹ But the current moment meets two basic conditions for regulation: (1) technological progress is causing harms that existing frameworks are poorly positioned to address, much as industrial production did a century ago, and (2) polls show that the public is more worried that the government will not do enough to regulate AI than that it will do too much.⁵⁰

Are we at a place where reasonable regulation sounds hopelessly ambitious? It is no more utopian than the pronouncements that frontier lab CEOs routinely make at international conferences. 

When social media first took hold in the 2000s, a failure to explore its negative effects, and an ill-advised deference to industry, resulted in lasting damage to society and politics. 

The stakes are even higher with AI. Let’s not make the same mistake twice.