A Deepfake of a Politician Nearly Swung an Indian Election. Nobody Was Punished.
A Deepfake of a Politician Nearly Swung an Indian Election. Nobody Was Punished.
By Akshay A. Walimbe
In April 2024, during the Lok Sabha election campaign, a video went viral across India. Home Minister Amit Shah appeared on screen, clearly and confidently saying that reservation for SC, ST and OBC communities is unconstitutional and that the BJP would repeal it.
In a country where reservation is one of the most emotionally charged political issues, the impact was immediate. The video exploded across WhatsApp groups and social media. Outrage followed. Accusations flew.
There was just one problem. Amit Shah never said it.
The video was a deepfake. In Shah’s original speech at a Telangana rally, he had said that the BJP would end unconstitutional Muslim reservation and redirect those rights to SCs, STs and OBCs. Someone had used AI to manipulate the footage, twisting his words into the exact opposite meaning.
Mumbai police registered a case against the Maharashtra Youth Congress’ social media handle and others, and at least six people were arrested in connection with the video’s circulation. But by then, the damage was done. The video had already reached millions. The correction never travels as fast as the lie.
Welcome to India’s first AI election.
Fifty Million Calls Nobody Made
The Amit Shah deepfake wasn’t an isolated incident. It was a symptom.
The 2024 Lok Sabha elections were widely dubbed the “AI election” and the numbers tell you why. According to GNET Research, in just the two months before polling began, over 50 million AI-generated voice clone calls were made to voters across India. Not automated robocalls with a tinny recorded message. AI-cloned voices that sounded like real politicians, having what felt like personal conversations.
One deepfake creator reportedly attempted to build enough synthetic audio and video data to convince 300,000 voters around Ajmer that they had each had a personalised conversation with a candidate. Parties reportedly spent an estimated $50 million on authorised AI-generated content during the campaign. Using AI voice calls was about 8 times cheaper than running a traditional call centre, according to multiple reports.
And that’s just the authorised stuff the content parties paid for openly. The unauthorised deepfakes were another matter entirely.
The Dead Come Back to Campaign
If the Amit Shah deepfake was about distortion, what happened in Tamil Nadu was about something even stranger: resurrection.
In January 2024, the DMK featured an AI-generated version of M. Karunanidhi at a campaign event. Karunanidhi the iconic Dravidian leader, recognisable by his trademark black sunglasses, white shirt, and yellow shawl had died in 2018. Yet there he was on screen, praising the DMK leadership, urging young cadres to fight for democracy, and lamenting that “many hard-fought states’ rights have been lost.”
According to Al Jazeera’s reporting, it was the third time in six months that the party had “resurrected” its former leader using AI. The three minute video was created by Muonium, an AI media tech firm. The DMK’s digital media wing confirmed it was their production.
Bollywood wasn’t spared either. Deepfakes of Ranveer Singh and Aamir Khan endorsing political parties circulated during the campaign. Both actors publicly denied any involvement. By the time the denial reached people, the endorsements had already done their work.
Deepfakes of Mamata Banerjee and a video showing Arvind Kejriwal singing inside jail also made the rounds. Fact checkers scrambled to keep up. Most voters never saw the fact checks.
The Problem Isn’t New. The Scale Is.
India actually has an earlier claim to political deepfake history. Back in February 2020 during the Delhi Assembly elections BJP MP Manoj Tiwari became, as MIT Technology Review reported, among the world’s first politicians to use deepfake technology for campaigning. His team created videos of him addressing voters in Hindi, Haryanvi, and English. Only the Hindi video was real. The other two used lip sync deepfake algorithms with a voice artist impersonating Tiwari.
Those videos were distributed across 5,800 WhatsApp groups in Delhi-NCR, reaching approximately 15 million people.
That was 2020. The technology was crude by today’s standards. It took effort, expertise, and time to produce a convincing deepfake. By 2024, the tools had become cheaper, faster, and accessible to anyone with a laptop and an internet connection.
The difference between 2020 and 2024 isn’t just quality. It’s volume. When 50 million AI voice calls and countless video deepfakes flood the information ecosystem during the world’s largest democratic exercise, we aren’t dealing with isolated incidents of media manipulation. We’re dealing with a structural crisis of truth.
The Liar’s Dividend
Here’s the part of the deepfake story that doesn’t get enough attention.
Law professors Bobby Chesney and Danielle Citron coined a term for it: the “liar’s dividend.” The idea is simple but devastating. As the public becomes more aware that deepfakes exist, a new weapon emerges: the ability to dismiss real content as fake.
Think about what that means for politics. A politician is caught on camera saying something damaging? “It’s a deepfake.” There’s audio evidence of a corrupt deal? “AI-generated.” Video evidence of police misconduct? “Can’t trust video anymore.”
The Amit Shah incident in India is a perfect illustration. A deepfake was made of things he didn’t say. But now, the existence of that deepfake also means that anything he actually does say on camera can be questioned. The deepfake didn’t just create a false reality it undermined the real one.
Real-world examples are already piling up globally. In the US, a lawyer for an alleged Capitol riot participant argued that prosecution video evidence was deepfaked. Tesla’s lawyers argued that Elon Musk’s past remarks on self driving safety should be dismissed because they “could be deepfakes.”
The liar’s dividend works in both directions. You can fabricate what never happened, and you can deny what actually did. Both weapons use the same ammunition: the erosion of trust.
Can’t We Just Detect Them?
You might think the answer is better detection technology. Build smarter AI to catch the fake AI. It sounds logical. It doesn’t work.
In laboratory conditions, state of the art deepfake detection systems achieve 94-96% accuracy. That sounds impressive. But against real world deepfakes the kind actually circulated on WhatsApp and social media performance drops by 45-50%, according to the Deepfake Eval 2024 benchmark. The best detection tools achieve about 50-65% accuracy in actual use.
That’s barely better than a coin flip.
Humans do even worse. Academic research shows we correctly identify high-quality deepfake videos only about 24.5% of the time. That means three out of four times, we can’t tell real from fake.
Detection is a losing arms race. Every time detection improves, the generation tools improve faster. The detector trained on yesterday’s deepfakes doesn’t recognise tomorrow’s.
Where Was the Regulator?
Here’s the timeline of India’s regulatory response to deepfakes in elections:
The Election Commission of India issued an advisory on May 6, 2024 after polling had already begun. It asked political parties to remove deepfake content within three hours of becoming aware of it. It asked parties to identify and warn individuals responsible for creating manipulated content.
It asked. It didn’t mandate. It didn’t enforce.
The ECI acted after the Delhi High Court ordered it to address the deepfake issue following a petition. Its approach was described by analysts as “largely depending on tech companies to self regulate.”
India’s broader regulatory evolution on deepfakes has been a series of advisories. MeitY issued its first advisory on deepfakes on November 7, 2023 triggered by the Rashmika Mandanna deepfake incident, where a 23 year old B-Tech graduate from Guntur, Andhra Pradesh, morphed the actress’s face onto a British Indian influencer’s video to gain Instagram followers. Subsequent advisories followed in December 2023 and March 2024, with draft amendments to IT Rules released in October 2025.
It wasn’t until February 10, 2026 that formal amendments to the IT Rules created a legal definition of “Synthetically Generated Information” and mandated platforms to remove nonconsensual AI-generated intimate imagery within two hours and other unlawful AI content within three hours.
That’s more than four years from the Manoj Tiwari deepfakes in 2020 to a formal legal definition. To be fair, India is not alone in struggling with this no major democracy has yet built an enforcement framework that can keep pace with deepfake technology. But four years is a long time when the technology is moving this fast, and India’s regulatory apparatus largely relied on platform self regulation during that window.
And the people who circulated the Amit Shah deepfake? At least six arrests were made and FIRs were filed. That is not nothing it shows law enforcement did act. But there has been no landmark prosecution resulting in conviction, no systemic deterrence, and no mechanism to prevent the same thing from happening in every state election, every municipal election, every panchayat election going forward.
The Provenance Crisis
The real solution isn’t detection it’s provenance. Instead of trying to figure out what’s fake, we need a way to prove what’s real.
The C2PA Coalition for Content Provenance and Authenticity is trying to build exactly this. Think of it as a “nutrition label for digital content,” showing where content originated, how it was created, and whether it’s been modified. As of 2025, the coalition includes over 3,700 members: Adobe, Microsoft, Google, OpenAI, Meta, Sony, Qualcomm, among others.
The concept is sound. But as of today, almost no internet content actually carries C2PA credentials. The technology exists; the adoption doesn’t. It faces a classic chicken and egg problem every camera, every phone, every editing tool needs to adopt the standard for it to work.
Meanwhile, according to an ORF analysis, 47% of Indian adults have either been victims of, or know someone who has been a victim of, an AI voice cloning or deepfake scam. That’s nearly double the global average of 25%. According to BW Businessworld, projected deepfake related financial losses in India could reach Rs 70,000 crore.
The 968 Million Voter Question
India has 968 million registered voters. It conducts elections at national, state, and local levels continuously throughout the year. It has 22 official languages, hundreds of dialects, and one of the highest WhatsApp penetration rates in the world a platform where content spreads virally and without editorial oversight.
It is, in other words, the perfect environment for deepfakes to thrive. And the regulatory infrastructure to deal with this reality is still being written in advisories, not in law; in suggestions, not in enforcement.
The 2024 election gave us a preview. Fifty million AI voice calls. Deepfakes of sitting ministers saying things they never said. Dead politicians brought back to life to campaign for their parties. Celebrity endorsements fabricated from whole cloth.
And this is not a partisan story. The BJP pioneered political deepfakes in India with Manoj Tiwari in 2020. The DMK resurrected Karunanidhi using AI. Congress linked accounts were arrested for circulating the Amit Shah deepfake. Deepfakes of opposition leaders like Mamata Banerjee and Arvind Kejriwal also circulated widely. Every political camp has used or been targeted by this technology. The enforcement response, across the board, has been inadequate. The technology got better. The next election will be worse.
In a world where you can’t tell real from fake, who decides what’s true?
I’m have written a book about exactly this how AI and automated systems make decisions about your life, where accountability disappears, and what we can do about it. If you want to know morea about this book or order a copy, you can do it here: https://akshaywalimbe.com/beyond-bias/