Intro
L1
·
Quiz
·
Lab
L2
·
Quiz
·
Lab
L3
·
Quiz
·
Lab
L4
·
Quiz
·
Lab
Module Test
Don't Get Fooled: AI and Lies · Introduction

Every New Medium Births a New Kind of Lie

Understanding the long history of fabricated reality — and why this moment is genuinely different.

In 1865, the same year the Civil War ended, American newspapers routinely ran fabricated photographs — composites stitched together in darkrooms that purported to show events that never occurred. By the 1890s, "yellow journalism" pioneers William Randolph Hearst and Joseph Pulitzer were commissioning illustrations of Spanish atrocities in Cuba that artists invented wholesale. The public, encountering photography and mass print for the first time, had no framework for systematic skepticism. Each new medium arrived faster than the public's ability to distrust it.

The same pattern replayed with recorded audio in the 1930s, broadcast television in the 1960s, desktop photo editing in the 1990s, and social-media video in the 2010s. In each era, a narrow window existed — sometimes only a few years — before audiences developed the instincts to question what they were seeing. We are currently inside that window for AI-generated media. Since 2022, tools capable of producing photorealistic images, convincing video, and cloned voices have moved from specialized research labs to free smartphone apps. The fabrications are now cheap, fast, and often visually indistinguishable from authentic footage.

This course is a practical guide to that window. It won't make you immune to manipulation — no course can — but it will give you a working vocabulary for what these tools are, documented cases of how they have already been used to deceive, and concrete detection habits you can apply immediately. The goal is calibrated skepticism: not paranoid rejection of all media, but the specific, informed suspicion that the moment requires.

Don't Get Fooled: AI and Lies · Lesson 1

The Video That Wasn't Real

How deepfakes work, where they came from, and the documented cases that put the technology on the global agenda.
What exactly is a deepfake — and how did a Reddit community accidentally build a geopolitical problem?

On January 1, 2019, Gabonese state television broadcast a video of President Ali Bongo Ondimba delivering a New Year's address to the nation. Bongo had been absent from public view for two months following a stroke in October 2018, and the video was intended to reassure citizens that he remained in command. Instead, it triggered a military coup attempt. Opposition politicians and Gabonese generals publicly declared the video fake — a deepfake, they argued, manufactured to disguise the president's true incapacitation. Within two weeks, army officers seized the national broadcaster in an attempted takeover. The coup failed, but the damage was done: a contested video, real or fabricated, had nearly toppled a government. Forensic analysts later disagreed about authenticity. The uncertainty itself was the weapon.

Where the Word "Deepfake" Came From

The term has a precise, traceable origin. In late 2017, a Reddit user posting under the handle "deepfakes" began sharing pornographic videos in which celebrities' faces had been digitally grafted onto performers' bodies. The account used a machine-learning technique called a Generative Adversarial Network, or GAN, which had been published as an academic paper by Ian Goodfellow and colleagues at the University of Montreal in 2014. Within weeks, other Reddit users had replicated the method. By January 2018, a developer had released an open-source desktop application called FakeApp that required no programming knowledge.

Reddit banned the deepfakes community in February 2018, citing its non-consensual pornography policy — but the code was already loose in the world. Within eighteen months, the same technique was being used to fabricate political speeches, financial fraud videos, and, as the Gabon case illustrated, potential instruments of destabilization. The technical democratization that took photography from a specialist craft to a smartphone default happened to deepfakes in roughly thirty-six months.

How the Technology Actually Works

A deepfake video is produced by a neural network trained on two things simultaneously: a large collection of images of a target person's face, and a source video of a different person performing the desired action. The GAN architecture pits two competing networks against each other — a generator that tries to produce convincing fake images, and a discriminator that tries to detect fakes. Over thousands of training iterations, the generator gets better at fooling the discriminator. The result is a face-swap that can track head pose, lighting, and expression in real time.

By 2022, the GAN approach had been largely supplanted by diffusion models — the technology underlying Stable Diffusion, Midjourney, and OpenAI's DALL-E. Diffusion models work differently: they learn to denoise randomly scrambled images, gradually recovering structure. They produce higher-quality outputs with less training data, which is why the quality ceiling rose so sharply between 2022 and 2024. Video synthesis tools including Runway ML (founded 2018, public release 2022) and Sora (OpenAI, announced February 2024) extend these capabilities to full motion video, including scenes with no source footage at all.

Why This Matters Beyond Celebrities

Early deepfake coverage focused heavily on celebrity face-swaps and political leaders. The more significant threat, documented by the FBI in a 2021 public service announcement, is targeted fraud: synthetic audio cloning a CEO's voice to authorize wire transfers, or fabricated video used in sextortion campaigns against private individuals. These low-profile, high-damage uses vastly outnumber the political cases in raw frequency.

Three Terms You Need

DeepfakeA synthetic media artifact — image, audio, or video — in which a person's likeness or voice has been generated or substantially altered by a neural network without their consent or knowledge, typically to make them appear to say or do something they did not.
GANGenerative Adversarial Network. A machine-learning architecture, described by Ian Goodfellow in 2014, in which two neural networks compete: a generator producing fakes and a discriminator evaluating them. The competition drives quality upward over training iterations.
Diffusion ModelA newer generation of generative AI that learns to reverse a noise-addition process, producing high-fidelity images or video from text prompts or reference material. Underlies most state-of-the-art image and video generators as of 2024.

The Detection Gap

Detection research has consistently lagged behind generation capability. In 2020, Facebook and Microsoft jointly released the Deepfake Detection Challenge dataset — over 100,000 videos — and offered prizes for detection algorithms. The winning model achieved 65% accuracy on the held-out test set. Human viewers tested on the same clips performed at roughly 50% — chance. More recent benchmarks using 2023-era synthetic video show detection rates falling further as generation quality improves.

This asymmetry has a practical consequence: you cannot reliably spot a high-quality deepfake by looking carefully. The visual tells that older guides described — unnatural blinking, blurry ear edges, inconsistent lighting on hair — have been largely engineered away. What remains useful is a set of contextual and behavioral habits: examining the source, verifying through independent channels, and applying particular suspicion to content that arrives already emotionally charged and demands immediate sharing.

Core Principle for This Module

Your eyes are no longer a reliable deepfake detector. Context, source verification, and behavioral skepticism are. This module will build those habits systematically across four lessons.

Lesson 1 Quiz

Four questions — select the best answer for each.
1. The term "deepfake" originated from which source in 2017?
Correct. The Reddit account "deepfakes" began posting AI face-swap pornography in late 2017, and the username became the name for the entire category of synthetic media.
Not quite. The term came from a Reddit username — "deepfakes" — who posted AI-generated pornographic videos in late 2017. Ian Goodfellow published GAN research in 2014 at the University of Montreal, not MIT, and the word wasn't his coinage.
2. In the January 2019 Gabon crisis, why did the disputed presidential video nearly cause a government collapse?
Correct. Forensic analysts disagreed about whether the video was authentic or fabricated. That unresolved uncertainty — not a confirmed fake — was enough to give military officers a pretext for a coup attempt.
Not quite. The video was never definitively proven fake — analysts disagreed. The lesson is that contested authenticity alone can serve as a political weapon, even without a confirmed fabrication.
3. What fundamental architecture underlies most state-of-the-art image and video generators as of 2024, having largely replaced GANs?
Correct. Diffusion models — which learn to reverse a noise process — now underlie tools like Stable Diffusion, Midjourney, DALL-E, and video generators like Runway and Sora, producing higher quality with less training data than GANs.
Not quite. Diffusion models have largely replaced GANs as the dominant architecture. They learn by reversing a noise-addition process and produce higher-quality outputs with less training data.
4. According to the 2020 Facebook/Microsoft Deepfake Detection Challenge results, what was the approximate accuracy of the winning detection algorithm?
Correct. The winning model hit 65% on the held-out test set. Human viewers on the same clips performed at roughly 50% — chance. This detection gap is precisely why visual inspection alone is an unreliable strategy.
Not quite. The winning algorithm achieved only about 65% accuracy, while human viewers performed at roughly 50% — essentially chance. The detection gap is real and has widened since 2020 as generation quality has improved.

Lab 1 — The Anatomy of a Deepfake

Conversation-based investigation · Minimum 3 exchanges to complete

Your Task

You're going to interrogate the AI assistant about how deepfake technology works, what its real-world limits are, and how the Gabon case fits into the broader history of synthetic media. Push back, ask follow-ups, and try to find the edges of what the technology can and cannot do.

Suggested opening: "Walk me through exactly how a GAN produces a face-swap — what is the generator actually doing step by step?" — or ask anything from Lesson 1 that you want to dig into further.
AI Lab Assistant
Deepfake Anatomy
Ready when you are. Ask me anything about how deepfakes are made, the history of the technology, the Gabon case, detection challenges, or the difference between GAN and diffusion-based approaches. I'll give you straight answers — including honest acknowledgment when something is genuinely uncertain.
Don't Get Fooled: AI and Lies · Lesson 2

Voice Cloning and the $25 Million Phone Call

Synthetic audio has already been used in large-scale financial fraud. Here is exactly how it happened.
How does voice cloning work — and why did a multinational corporation wire $25 million to fraudsters based on a single video call?

In February 2024, a finance worker at the Hong Kong branch of a British multinational engineering firm attended what appeared to be a routine video conference call with his company's UK-based Chief Financial Officer and several other colleagues. The CFO instructed him to execute a series of wire transfers totaling HK$200 million — approximately US$25.6 million. The worker harbored initial doubts but was reassured by seeing and hearing the CFO and other familiar colleagues on screen. He made fifteen separate transactions. Every person on that call except the worker himself was an AI-generated avatar. Hong Kong police confirmed the fraud in February 2024 and described it as the largest deepfake-assisted financial crime recorded to that point.

How Voice Cloning Works

Modern voice synthesis requires remarkably little source material. ElevenLabs, a startup founded in 2022 and valued at over $1 billion by 2024, offers cloning from as little as one minute of audio. Its underlying technology — a neural text-to-speech model trained on massive audio corpora — learns the spectral fingerprint of a voice: pitch distribution, formant patterns, speaking rhythm, and prosody. Given a text string, it then synthesizes novel speech that sounds like the target speaker saying words they never recorded.

The technical barrier fell sharply in 2023. Microsoft's VALL-E, demonstrated in January 2023, could clone a voice from three seconds of audio. The same month, Meta released Voicebox, capable of in-context speech synthesis across six languages. These tools were research demonstrations, but the underlying weights often leaked or inspired open-source reimplementations within months of publication.

Documented Fraud Cases Before Hong Kong

The Hong Kong case had predecessors. In March 2019, the CEO of a UK-based energy company received a phone call from his parent company's German executive — or so he believed. The voice instructed him to wire €220,000 to a Hungarian supplier within the hour. The CEO complied. The Wall Street Journal reported this case in August 2019; the "German executive" was a voice cloning system. It was the first publicly documented corporate AI voice fraud.

In 2021 and 2022, the FBI's Internet Crime Complaint Center began receiving reports of "virtual kidnapping" scams in which a family member's cloned voice was used to simulate distress calls demanding ransom. By 2023, these scams had been documented in at least four US states. The Arizona mother Jennifer DeStefano testified before the US Senate in June 2023 about receiving a call in which she heard what she was certain was her daughter's voice screaming — it was synthesized from publicly available social media videos.

The Liveness Problem

Security systems that rely on voice authentication — phone banking, corporate verification calls — assume a live speaker. Cloned voices defeat these systems entirely. HSBC, Barclays, and other major banks deployed voice ID as a security layer in the mid-2010s; the assumption that voice is biometrically difficult to fake is now outdated. Several banks began phasing out voice biometrics as a standalone authentication factor in 2023.

Why It Worked: The Cognitive Mechanics

The Hong Kong fraud succeeded not because the technology was flawless but because it exploited specific cognitive vulnerabilities. The finance worker reported being "reassured" by seeing multiple familiar faces simultaneously — a social proof effect. The call mimicked a routine business context with familiar participants, reducing the cognitive load that would otherwise trigger skepticism. Urgency was manufactured through the framing of a time-sensitive transfer.

Psychologists call this contextual coherence bias — when a situation's overall shape matches our expectations, we reduce scrutiny of individual components. Fraudsters understood this before AI gave them the tools to exploit it technically. The technology did not create a new psychological vulnerability; it created a new and highly efficient method of exploiting an existing one.

Voice CloningNeural synthesis of speech that replicates a specific person's voice characteristics — pitch, timbre, rhythm, accent — from a reference audio sample, allowing arbitrary text to be spoken in that voice without the person's participation.
Contextual Coherence BiasA cognitive tendency to reduce scrutiny of individual elements when the overall context of a situation matches prior expectations. Exploited by fraud scenarios that replicate familiar settings.
Practical Defense

Financial institutions and security researchers now recommend "callback verification" — independently calling a known number for the person who made the request, never using contact information provided during the suspicious communication itself. This low-tech habit defeats virtually all voice-cloning fraud scenarios, regardless of audio quality.

Lesson 2 Quiz

Four questions on voice cloning and audio fraud.
1. The February 2024 Hong Kong deepfake fraud resulted in a transfer of approximately how much money?
Correct. The worker made fifteen wire transfers totaling HK$200 million — approximately US$25.6 million — after attending a video call in which every other participant was AI-generated.
Not quite. The amount was HK$200 million, approximately US$25.6 million. The €220,000 figure is from the earlier 2019 UK energy company fraud, which used voice-only cloning.
2. Microsoft's VALL-E (January 2023) demonstrated voice cloning from what minimum audio sample length?
Correct. VALL-E, demonstrated in January 2023, required only three seconds of reference audio — a threshold that makes virtually any phone call, online video, or podcast appearance a viable source for cloning.
Not quite. VALL-E needed only three seconds. ElevenLabs requires about one minute. The technological threshold has dropped far below what most people assume.
3. What is "contextual coherence bias" as described in Lesson 2?
Correct. When a situation's overall shape — familiar colleagues, routine context, expected format — matches our expectations, we reduce scrutiny of individual elements. Deepfake fraudsters exploit this by engineering convincing contexts, not just convincing faces.
Not quite. Contextual coherence bias refers to reduced individual scrutiny when the overall situation seems familiar and expected. It's why the Hong Kong worker believed the call even though he initially had doubts — the overall scene was too normal-seeming.
4. What low-tech defense do security researchers recommend against voice-cloning fraud?
Correct. "Callback verification" — hanging up and dialing a number you already have on record — defeats voice-cloning fraud regardless of audio quality, because it takes communication out of the channel the fraudster controls.
Not quite. Asking a personal question doesn't work if the fraudster has prepared answers from your social media. The reliable defense is callback verification: independently contacting the requester via a channel and number you already control.

Lab 2 — Voice Fraud Scenarios

Conversation-based investigation · Minimum 3 exchanges to complete

Your Task

You'll work through real and hypothetical voice-cloning fraud scenarios with the AI assistant. Try to identify the specific moment a target could have broken the chain of deception in each case. Ask about technical limits, organizational defenses, and why voice authentication is now considered a weakened security layer.

Suggested opening: "In the 2019 UK energy company case, what was the single most exploitable moment — where could the CEO have broken the fraud?" — or design your own line of questioning.
AI Lab Assistant
Voice Cloning & Fraud
Let's dig into voice cloning fraud. I can walk you through the documented cases — the 2019 UK energy CEO, the 2023 Jennifer DeStefano virtual kidnapping, the 2024 Hong Kong video call — and help you analyze where defenses broke down. Ask me anything.
Don't Get Fooled: AI and Lies · Lesson 3

AI-Generated Images in the News

Fabricated photographs have already shaped public perception of real events. These are the cases that made newsrooms reckon with AI.
When AI-generated images spread through news cycles, what damage do they actually cause — and how have institutions responded?

On the morning of May 22, 2023, an image circulated rapidly across Twitter and Facebook showing a large explosion near the Pentagon in Arlington, Virginia. The image was photorealistic — a billowing smoke plume, architectural details consistent with the building, dramatic lighting. Within minutes it had been shared by verified accounts including BloombergFeed, a parody account that at the time carried a blue verification checkmark. The S&P 500 dipped briefly before markets realized no explosion had occurred. The Arlington Fire Department and Pentagon press office issued denials. The image was synthetic — generated by an AI image tool. It was, to date, one of the most consequential single AI-fabricated images in terms of immediate, measurable financial impact.

The Arrested Trump Images — March 2023

Two weeks before the Pentagon image, in March 2023, Eliot Higgins — founder of the open-source intelligence organization Bellingcat — published a series of AI-generated images depicting Donald Trump being arrested and physically restrained by police officers. Higgins created the images using Midjourney and explicitly labeled them as synthetic, posting them as an experiment in what the technology could produce. They spread anyway, stripped of context, across Telegram channels, WhatsApp groups, and right-wing news aggregators. Midjourney subsequently banned Higgins from the platform, citing its policies on generating images of real political figures. The episode illustrated a structural problem: explicit labeling does not prevent decontextualization at scale.

The Pope's Puffer Jacket — March 2023

In the same month, an image of Pope Francis wearing a white designer puffer jacket went globally viral. Millions of people shared it as genuine. The image was created by a Chicago construction worker named Pablo Xavier using Midjourney v5; he told BuzzFeed News he had generated it while under the influence of a mild psychedelic and was surprised it spread so far. The image was eventually traced and debunked by fact-checkers at Reuters and AFP, but only after it had achieved wide genuine belief. This case is significant because it involved no political motive or organized campaign — a single person, a consumer AI tool, and an unexpected viral moment.

The Pope image made visible something researchers had been documenting more quietly: AI image generation had crossed a realism threshold at which ordinary people's prior confidence in photographic authenticity — developed over 185 years of photography — could not be relied upon. The aesthetic tells that had distinguished CGI from photography for decades were gone.

The Journalism Response

Following the Pentagon image and similar incidents, the Associated Press updated its AI use policy in July 2023, prohibiting the use of AI-generated images in editorial content and requiring byline disclosure for any AI-assisted work. Reuters and Agence France-Presse issued similar policies. The News/Media Alliance, representing 2,000 US publishers, published guidance on AI image verification protocols. These policies represent the journalism industry's first systematic response to synthetic image proliferation — but they govern professional newsrooms, not the social media accounts that now originate most viral content.

How to Actually Evaluate a Suspicious Image

Reverse image search remains the first-line tool. Google Lens, TinEye, and Yandex Images can identify if an image has appeared before in different contexts. A novel image with no prior appearance history appearing during a breaking news event warrants heightened suspicion.

Metadata examination is useful for images that haven't been stripped. Genuine camera images embed EXIF data — camera model, GPS coordinates, timestamp. Images generated by AI tools typically lack this metadata entirely, or contain metadata inconsistent with the claimed context.

AI detection tools exist — Hive Moderation, Illuminarty, and Google's SynthID (applied to images generated by Google's own tools) — but these have significant false-positive and false-negative rates. They are useful as a single data point, not a verdict. The most reliable detection approach remains a combination of source verification, contextual implausibility assessment, and reverse-search cross-referencing.

DecontextualizationThe process by which content is stripped of its original framing — including labels, caveats, and source information — as it spreads across platforms, allowing synthetic or satirical material to be received as factual by downstream audiences.
EXIF DataExchangeable Image File Format metadata embedded by digital cameras in photograph files, including camera model, lens, aperture, shutter speed, GPS coordinates, and timestamp. AI-generated images typically lack authentic EXIF data.
The Core Diagnostic Question

Before sharing any image from a breaking news situation: Where did this specific file come from, and who is the first person claiming it is real? If you cannot answer both questions, do not share. This rule, applied consistently, eliminates a substantial fraction of synthetic image spread before any technical tool is needed.

Lesson 3 Quiz

Four questions on AI-generated images and their real-world impact.
1. What measurable real-world impact did the fake Pentagon explosion image have on May 22, 2023?
Correct. The S&P 500 dipped briefly as markets reacted to the image before the Arlington Fire Department and Pentagon confirmed no explosion had occurred. This is one of the clearest documented cases of a single synthetic image causing financial market movement.
Not quite. The image caused a measurable — if brief — dip in the S&P 500. It was amplified by the BloombergFeed parody account, which carried a Twitter blue checkmark at the time, adding apparent credibility.
2. Eliot Higgins's March 2023 Trump arrest images spread despite being explicitly labeled as AI-generated. What structural problem does this illustrate?
Correct. Higgins labeled his images clearly as AI-generated experiments. They spread anyway through Telegram and WhatsApp stripped of that context. Labeling protects original audiences; it cannot control what downstream sharers do with the content.
Not quite. The core problem is decontextualization: images can be copied, screenshotted, or re-shared without their original labels. Clear labeling protects the first audience but not subsequent ones at scale.
3. Who created the viral Pope Francis puffer jacket image, and what tool did they use?
Correct. Pablo Xavier, a Chicago construction worker with no political motive, created the image using Midjourney v5 and was reportedly surprised by how far it spread. The case shows that consequential synthetic images don't require organized campaigns or sophisticated actors.
Not quite. It was Pablo Xavier, a Chicago construction worker, using Midjourney v5. He told BuzzFeed News he created it recreationally and was surprised by the spread. No political motive was involved.
4. What does the absence of EXIF data in an image file suggest — and why is this only a single data point rather than a verdict?
Correct. Many platforms — including Twitter/X, Facebook, and WhatsApp — strip EXIF data from images automatically during upload. Missing metadata is suspicious but not conclusive. It should be combined with reverse image search results, contextual plausibility, and other signals.
Not quite. While AI-generated images do typically lack authentic EXIF data, most social platforms strip metadata from all uploaded images regardless of origin. Missing EXIF raises a flag but is not by itself diagnostic.

Lab 3 — Image Verification Methods

Conversation-based investigation · Minimum 3 exchanges to complete

Your Task

Work through a systematic image verification workflow with the AI assistant. You'll examine specific scenarios — a breaking news image, a social media viral photo, a historical document scan — and build a step-by-step verification protocol you could actually apply. Challenge the assistant on edge cases where the standard tools fail.

Suggested opening: "Walk me through exactly what you would do in the first five minutes after seeing a dramatic image arrive on Twitter during a claimed breaking news event — step by step, in order of what to check first." — or ask your own question.
AI Lab Assistant
Image Verification
I'm ready to work through image verification with you. I can cover reverse image search techniques, EXIF metadata analysis, AI detection tool limitations, the specific documented cases from Lesson 3, or help you build a verification protocol for specific scenarios you have in mind.
Don't Get Fooled: AI and Lies · Lesson 4

The Liar's Dividend

The most dangerous effect of deepfakes may not be the fakes themselves — it is the plausible deniability they give to authentic evidence.
How does the existence of deepfake technology allow real wrongdoing to be dismissed as fabricated — and what documented cases have already exploited this?

In 2018, Facebook's platform became the primary distribution channel for anti-Rohingya propaganda in Myanmar, contributing to conditions that the United Nations would later describe as constituting genocide. Authentic video evidence of military atrocities — real footage, not synthetic — was systematically dismissed by military officials and their supporters as "fabricated" and "fake news." The dismissal worked partly because audiences had already been primed by years of discussion about digital manipulation. The existence of a category called "fake video" gave bad actors a ready-made defense against authentic documentation. Researchers Nina Schick and Danielle Citron, who documented this dynamic in 2019 and 2020 respectively, named it the Liar's Dividend: the phenomenon by which the mere existence of synthetic media allows real media to be dismissed.

The Concept Defined

The term "Liar's Dividend" was coined by law professors Bobby Chesney and Danielle Citron in their 2019 paper "Deep Fakes: A Looming Crisis for National Security, Democracy, and Privacy" (published in the California Law Review). Their core argument was counterintuitive: the primary damage from deepfake technology would not be from people believing false things, but from people acquiring a justifiable excuse to disbelieve true things.

The asymmetry matters. To fool someone with a deepfake requires a high-quality fabrication, a plausible distribution story, and some luck. To exploit the Liar's Dividend requires only a claim — "that could be deepfaked" — delivered into an already polarized information environment. The defensive move is trivially easy; the authentic documentation it undermines is often irreplaceable.

Documented Cases of the Liar's Dividend

In April 2021, video emerged appearing to show Gabonese soldiers committing summary executions of civilians. The Gabonese government immediately declared the video a deepfake. Independent verification by Amnesty International's Digital Verification Corps and Bellingcat ultimately concluded the footage was authentic. But the government's "deepfake" claim occupied media attention for weeks and reduced the video's immediate political impact internationally.

In December 2022, a US congressman under investigation for financial misconduct publicly suggested that incriminating text messages and emails presented against him could have been "AI-generated fabrications." No evidence supported this claim. The suggestion alone — requiring no proof — was sufficient to introduce uncertainty in some media coverage. Chesney and Citron had predicted this exact dynamic in 2019.

In 2023, multiple defendants in legal proceedings in the United States raised "deepfake defenses" — claiming that video or audio evidence submitted against them could be synthetic. Courts have so far generally rejected these claims when prosecutors can demonstrate chain of custody, but the proceedings have become more expensive and time-consuming as a result. Legal scholars predict this will become standard practice regardless of actual evidence authenticity.

The Epistemic Corrosion Problem

Chesney and Citron argued that the deepfake era risks producing what they called "epistemic corrosion" — a generalized collapse of trust in audiovisual evidence that advantages those in power (who have the resources to fight evidence) and disadvantages those with less power (journalists, human rights workers, ordinary citizens attempting accountability). The issue is not just individual gullibility — it is the structural damage to shared evidentiary standards.

Responses and Partial Solutions

Content credentials and provenance standards represent the most serious technical response. The Coalition for Content Provenance and Authenticity (C2PA), formed in 2021 and backed by Adobe, Microsoft, Intel, the BBC, and the Associated Press, has developed an open standard for cryptographically signing media at the point of creation. A C2PA-compliant camera embeds a cryptographic certificate at capture; any editing is logged and signed. The New York Times adopted C2PA credentials for its photojournalism in 2023.

Legal frameworks are still developing. The DEEPFAKES Accountability Act was introduced in the US Congress in 2019 and again in 2023 but had not been enacted as of early 2024. Several states — including California (AB 730, 2019) and Texas (SB 751, 2023) — passed laws targeting deepfakes in electoral contexts. The EU's AI Act, approved by the European Parliament in March 2024, includes disclosure requirements for AI-generated content.

None of these responses fully addresses the Liar's Dividend, because the problem is not primarily technical. A camera with cryptographic provenance cannot retroactively credential the footage a human rights worker shot on a phone in 2018. The asymmetry between producing doubt and producing trust remains structurally unfavorable.

Liar's DividendA term coined by law professors Bobby Chesney and Danielle Citron (2019) describing the strategic benefit that the existence of deepfake technology confers on anyone wishing to dismiss authentic audiovisual evidence as fabricated — requiring no proof, only the claim of possibility.
C2PACoalition for Content Provenance and Authenticity. An industry body formed in 2021 that has developed an open standard for cryptographically signing media at the point of creation, creating a verifiable chain of custody for images and video.
The Takeaway for This Module

Calibrated skepticism cuts in both directions. The same critical habits that protect you from believing fabricated content should also protect you from dismissing authentic content on the basis of unsubstantiated deepfake claims. Ask for the affirmative evidence of fabrication, not just the assertion of possibility. "This could be fake" is not the same as "this is fake."

Lesson 4 Quiz

Four questions on the Liar's Dividend and institutional responses.
1. Who coined the term "Liar's Dividend" and in what publication?
Correct. Bobby Chesney and Danielle Citron coined the term in their paper "Deep Fakes: A Looming Crisis for National Security, Democracy, and Privacy," published in the California Law Review in 2019.
Not quite. The term was coined by law professors Bobby Chesney and Danielle Citron in a 2019 California Law Review paper. Nina Schick did write about deepfakes, but the Liar's Dividend terminology originates with Chesney and Citron.
2. What makes the Liar's Dividend asymmetric and therefore particularly dangerous?
Correct. The asymmetry is the core of the concept: saying "that could be a deepfake" is effortless and requires no evidence. Proving authenticity requires chain-of-custody documentation, technical analysis, and often significant resources. Bad actors get an asymmetric advantage.
Not quite. The asymmetry Chesney and Citron identified is between claiming possible fabrication (effortless, requires no proof) and establishing authenticity (requires documentation, technical analysis, and resources). The doubt is cheap; the rebuttal is expensive.
3. What does the C2PA standard do, and which organizations founded it?
Correct. C2PA — the Coalition for Content Provenance and Authenticity — was formed in 2021 by Adobe, Microsoft, Intel, the BBC, and the Associated Press, among others. It creates an open standard for cryptographic signing of media at creation, making the editing history verifiable.
Not quite. C2PA is an industry coalition — not a government program — formed in 2021 by Adobe, Microsoft, Intel, the BBC, and the AP. It cryptographically signs media at the point of capture or creation, creating a verifiable provenance chain.
4. According to Lesson 4, what is the correct epistemic response when someone claims a piece of video evidence "could be a deepfake"?
Correct. "This could be fake" is not the same as "this is fake." Calibrated skepticism requires demanding affirmative evidence of fabrication, not merely the assertion that fabrication is technically possible. Otherwise you are exploitable through the Liar's Dividend.
Not quite. The correct response is to ask what positive evidence supports the fabrication claim. Possibility is not proof. If you treat every unverified video as fake pending C2PA certification, you have handed bad actors a perfect tool for burying authentic evidence.

Lab 4 — The Liar's Dividend in Practice

Conversation-based investigation · Minimum 3 exchanges to complete

Your Task

Explore the Liar's Dividend with the AI assistant. Work through specific scenarios in which authentic evidence has been or could be dismissed as AI-generated. Try to identify which institutional defenses — C2PA, legal chain-of-custody standards, journalistic verification protocols — actually address the problem and which ones don't reach it.

Suggested opening: "Can C2PA actually solve the Liar's Dividend problem — or does it only work for future content captured on compliant devices? What happens to all the authentic footage that already exists without cryptographic provenance?" — or probe the problem from a different angle.
AI Lab Assistant
Liar's Dividend
Let's examine the Liar's Dividend carefully. I can discuss the Chesney-Citron framework, walk through the Myanmar, Gabon, and US legal cases in more detail, analyze the limits of C2PA and legal chain-of-custody, or help you think through scenarios where the Liar's Dividend has already been deployed. Where do you want to start?

Module 1 Test

15 questions across all four lessons — 80% required to pass.
1. The GAN architecture that underlies early deepfakes was first described in a paper by Ian Goodfellow in what year and at which institution?
Correct. Ian Goodfellow published the GAN paper in 2014 at the University of Montreal. It was this research that the Reddit deepfakes community weaponized in 2017.
The GAN paper was published in 2014 at the University of Montreal by Ian Goodfellow and colleagues.
2. FakeApp, the open-source tool that democratized face-swap video creation, was released in which month and year?
Correct. FakeApp was released in January 2018, shortly before Reddit banned the deepfakes community in February 2018. By then the code was already widely distributed.
FakeApp was released in January 2018. Reddit banned the deepfakes subreddit in February 2018 — but the software was already out.
3. In the 2019 UK energy company voice cloning fraud, how much money was transferred and to where?
Correct. The UK energy CEO was instructed by a cloned voice purporting to be his parent company's German executive to wire €220,000 to a Hungarian supplier. The Wall Street Journal reported this in August 2019.
The amount was €220,000 to a Hungarian supplier — the first publicly documented corporate AI voice fraud case, reported by the Wall Street Journal in August 2019.
4. Jennifer DeStefano testified before the US Senate about a virtual kidnapping scam. What was the source of the cloned voice?
Correct. The fraudsters synthesized the daughter's voice from publicly available social media videos — no special access was required. This illustrates how any public audio presence creates cloning vulnerability.
The voice was cloned from publicly available social media videos of her daughter — no hacking or special access was needed. Any public audio or video is now a potential voice cloning source.
5. The 2020 Deepfake Detection Challenge was jointly produced by which two technology companies?
Correct. Facebook and Microsoft jointly released the Deepfake Detection Challenge dataset of over 100,000 videos. The winning detection algorithm achieved approximately 65% accuracy.
The challenge was produced by Facebook and Microsoft, with over 100,000 videos. The best detection algorithm reached only 65% accuracy on the test set.
6. OpenAI's Sora video generation system was publicly announced in what month and year?
Correct. Sora was announced in February 2024, capable of generating high-fidelity video from text prompts — including scenes with no source footage at all.
Sora was announced by OpenAI in February 2024.
7. The fake Pentagon explosion image on May 22, 2023 was amplified by a parody account called BloombergFeed. What gave it apparent credibility?
Correct. After Twitter's ownership change in 2022, verification checkmarks became purchasable, meaning parody accounts could carry the same visual credibility signal as genuine news organizations — a structural vulnerability the Pentagon image exploited.
BloombergFeed carried a Twitter blue verification checkmark — which had become purchasable after the 2022 ownership change — lending it false credibility as a news source.
8. Eliot Higgins, who created the AI Trump arrest images in March 2023, is also the founder of which organization?
Correct. Eliot Higgins founded Bellingcat, the open-source intelligence organization. He created the Midjourney images explicitly as a demonstration of what the technology could produce — and was subsequently banned from the platform.
Higgins founded Bellingcat, the open-source intelligence organization known for, among other things, its investigation of the MH17 shootdown. He was banned from Midjourney following the Trump images.
9. What is the significance of the Pope Francis puffer jacket case for understanding AI image risk?
Correct. Pablo Xavier had no political motive — he created the image recreationally and was surprised it spread. The case reveals that high-impact synthetic image proliferation doesn't require adversarial intent or organizational resources.
The key lesson is the absence of intent: a single person with a consumer AI tool and no agenda created one of 2023's most viral synthetic images. Risk doesn't require malicious actors.
10. California's AB 730, passed in 2019, addressed deepfakes in what specific context?
Correct. California AB 730 targeted deepfakes specifically in electoral contexts, prohibiting the distribution of materially deceptive synthetic media depicting candidates within 60 days of an election.
AB 730 specifically addressed deepfakes in electoral contexts — materially deceptive synthetic media of candidates within 60 days of an election. Separate California legislation addressed non-consensual intimate deepfakes.
11. The "Liar's Dividend" as described by Chesney and Citron primarily damages which group most?
Correct. The Liar's Dividend structurally advantages those in power who want to suppress evidence and disadvantages those attempting accountability. Fighting a "deepfake" claim is expensive; making it is free.
Chesney and Citron argued the structural damage falls hardest on those seeking accountability — journalists, human rights workers, ordinary citizens — whose authentic documentation can be dismissed with a costless "deepfake" claim.
12. What does "callback verification" defend against, and why is it effective regardless of audio quality?
Correct. Callback verification works because it takes the interaction out of the fraudster's controlled channel. However realistic the cloned voice, the fraudster cannot intercept or answer a call to a number you independently possess and dial yourself.
Callback verification defeats voice cloning not by detecting fakes but by exiting the fraudster's communication channel entirely. No matter how good the voice clone, the fraudster cannot answer a call you make to a number you already have on independent record.
13. Which of the following best describes a "diffusion model" as used in AI image generation?
Correct. Diffusion models learn by training on progressively noisier versions of images and learning to reverse the process. This approach produces higher-quality results with less training data than GANs and underlies most state-of-the-art image generators as of 2024.
Diffusion models learn to reverse a noise-addition process — they train on progressively scrambled images and learn to recover the original structure, enabling generation of novel high-fidelity images from text descriptions.
14. The C2PA standard was adopted for photojournalism by which major news organization in 2023?
Correct. The New York Times adopted C2PA credentials for its photojournalism in 2023, embedding cryptographic provenance in images at the point of capture.
The New York Times adopted C2PA credentials for its photojournalism in 2023. The AP and Reuters were among the founding members of the C2PA coalition but the NYT adoption was the specific 2023 implementation noted in the lesson.
15. A person claims that video evidence against them in a legal proceeding is a deepfake — but offers no technical evidence of this. According to this module's framework, what is the correct response?
Correct. Calibrated skepticism cuts both ways. "This could be a deepfake" is not proof of fabrication. Chain-of-custody documentation and technical forensics establish authenticity; the mere possibility of fakery does not rebut them without affirmative evidence.
The module's core principle is that calibrated skepticism applies in both directions. Demanding proof of fabrication — not just assertion of possibility — is the correct epistemic response. Chain-of-custody documentation addresses authenticity questions independently of whether deepfakes exist as a technology category.