Deep Fakes Are a Looming Threat to Privacy and Democracy

Deep Fakes Are a Looming Threat to Privacy and Democracy

Imagine watching a video of the President of the United States declaring nuclear war. The lighting is right. The voice is perfect. The facial tics are familiar. You believe it. Millions of people believe it. And then you learn the whole thing was generated by a machine learning algorithm trained on thousands of hours of real footage. You are not watching reality. You are watching a statistical hallucination.

That is not a hypothetical future. That is what the legal scholars Robert Chesney and Danielle Keats Citron call a "deep fake," and their 2018 paper "Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security" is the first serious attempt to map the full scale of the threat (Chesney & Citron, 2018). The paper is terrifying not because it predicts some distant dystopia, but because it shows how the technology is already here, and our legal and social defenses are not.

Why This Is Different From Old School Lies

Human beings have always lied. Propaganda is older than printing presses. But deep fakes are not just better lies. They are a fundamentally different kind of threat.

Chesney and Citron argue that deep fakes exploit something deeper than our trust in media. They exploit our cognitive architecture. Our brains evolved to trust what we see and hear. A video of a person saying something feels more real than a written quote. That is why television news has always been more powerful than newspapers. Deep fakes weaponize that instinct.

The authors break down the danger into three categories: privacy, democracy, and national security. Each one is bad on its own. Together, they create a perfect storm.

Privacy: When Your Face Becomes a Weapon

The most immediate harm is to individuals. Chesney and Citron describe a world where anyone can be inserted into a pornographic video, or made to say racist things, or appear to confess to a crime. The technology does not require Hollywood budgets. Open source tools exist. A person with basic coding skills and a laptop can generate convincing fake audio and video.

The authors point to a specific example that predates their paper: in 2017, a Reddit user named "deepfakes" began posting fake celebrity porn videos. The technology spread rapidly. Within months, nonconsensual deep fake pornography was being used to harass and humiliate women, often by ex-partners or online trolls. The legal system had no framework for handling this. Revenge porn laws were designed for real images. Deep fakes existed in a gray zone.

Chesney and Citron note that the privacy harm goes beyond embarrassment. Deep fakes can be used for "personal sabotage" — a fake video of a teacher saying something racist could get them fired. A fake audio recording of a business executive discussing illegal activity could tank a company's stock price. The authors found that existing defamation law is poorly equipped to handle these cases, because the damages are so diffuse and the perpetrators are often anonymous.

Democracy: The End of the "Marketplace of Ideas"

This is where the paper gets genuinely disturbing. The authors describe a concept called "truth decay" — the erosion of shared facts as a basis for public debate. They argue that deep fakes will supercharge this process.

Here is the mechanism. In a healthy democracy, citizens need to agree on basic facts. Did the candidate say that? Did the protest turn violent? Was that video edited? Deep fakes make it possible to deny any inconvenient truth by claiming it is a fake. This is called the "liar's dividend." If you are a politician caught on tape saying something awful, you can simply claim the tape is a deep fake. Even if experts prove it is real, the damage is done. The doubt lingers.

Chesney and Citron cite evidence from political science showing that people are more likely to believe misinformation that confirms their existing biases. Deep fakes exploit this. A fake video of a political opponent saying something outrageous will spread like wildfire through social media, because it confirms what people already want to believe. By the time fact checkers debunk it, the election might be over.

The authors also warn about "information cascades" — a phenomenon where false information spreads so quickly that it creates its own reality. In 2016, fake news stories outperformed real news stories on Facebook. Deep fakes will make that problem exponentially worse, because video is harder to debunk than text.

National Security: The Covert Action Nightmare

The most alarming section of the paper deals with national security. Chesney and Citron describe scenarios that read like a Tom Clancy novel, but with a chilling footnote: this technology already exists.

Imagine a deep fake video of a foreign leader appearing to declare war. Or a fake audio recording of a general ordering a strike. The authors note that deep fakes could be used for "strategic deception" — planting false evidence to trigger a military response, or to discredit intelligence agencies. In a crisis, speed matters. A deep fake could cause irreversible damage before anyone verifies it.

The paper also discusses the problem of "blackmail at scale." Foreign intelligence agencies could generate deep fake pornographic videos of government officials, then threaten to release them unless the official cooperates. The official might be innocent, but the threat of public humiliation is enough. The authors found that existing counterintelligence frameworks do not account for this form of coercion.

What Can Actually Be Done?

Chesney and Citron do not just describe the problem. They spend a significant portion of the paper evaluating solutions. The picture is not pretty.

Technological Solutions

The authors examine detection tools — algorithms designed to spot deep fakes by looking for artifacts like inconsistent lighting or unnatural eye movements. The problem is that detection is an arms race. As detection improves, generation improves. The authors found that by 2018, deep fakes were already becoming "increasingly resistant to detection." The gap between generation and detection is likely to widen, not close.

Legal Solutions

Chesney and Citron propose criminal penalties for malicious deep fakes, similar to laws against revenge porn. They also explore civil liability — allowing victims to sue for defamation or intentional infliction of emotional distress. But they are honest about the limitations. Laws are slow. Technology is fast. And the First Amendment protects a lot of speech, even false speech.

The authors also discuss the role of platforms like Facebook and YouTube. They note that Section 230 of the Communications Decency Act gives platforms broad immunity for user-generated content. This means platforms have little legal incentive to police deep fakes. The authors recommend reforming Section 230 to create liability for platforms that knowingly host malicious deep fakes.

Market Solutions

The paper explores the idea of "authentication trails" — cryptographic signatures that prove a video was recorded by a specific device at a specific time. If all cameras and microphones automatically signed their output, deep fakes would be easier to spot. But the authors note that this would require massive infrastructure changes and would not prevent deep fakes created from older, unsigned footage.

What the Research Does Not Prove

Chesney and Citron are careful to note what they do not know. The paper is a framework, not a prediction. They do not claim that deep fakes will destroy democracy. They argue that the risk is real and significant, but they also acknowledge that society has adapted to previous technological disruptions. The printing press created propaganda. Radio created demagogues. The internet created fake news. Each time, we found ways to cope.

The open question is whether deep fakes are different in kind or just in degree. The authors lean toward "different in kind," because of the unique power of video to bypass rational thought. But they do not have data to prove this. The paper is a warning, not a prophecy.

What This Actually Means

• ▸If you are a public figure, assume a deep fake of you exists. Politicians, journalists, and activists should have a response plan. Denial is not enough. You need a system for rapid authentication, like a trusted third party that can verify your real statements.
• ▸Do not trust video evidence from an election cycle. If a video surfaces of a candidate saying something shocking, assume it could be fake until verified by multiple independent sources. This is not cynicism. It is survival.
• ▸Platforms need to be held accountable now. Waiting for perfect detection technology is a mistake. The authors recommend that platforms should be required to watermark AI-generated content and to take down malicious deep fakes within hours, not days.
• ▸Laws need to catch up to the technology. Criminal penalties for nonconsensual deep fake pornography and for deep fakes intended to influence elections are a minimum. The First Amendment is not a suicide pact.
• ▸The liar's dividend is already being used. Any time a politician claims a real recording is a deep fake, they are exploiting the technology even if no deep fake was used. The mere existence of the technology creates plausible deniability. This is the most insidious harm of all.

Chesney and Citron end their paper with a call for action. They write that "the time to act is now, before the technology becomes so widespread that the damage is irreversible." That was 2018. The clock is still ticking.