Kartik Sahu, a 60-something private car driver, spends hours on Facebook on his smartphone. When he is not watching trailers of Bengali films, he is swiping content related to politics. A lot of it is “news” from individual content creators. If Monday finds Suvendu Adhikari or Samik Bhattacharya making five promises to sway his mood, Tuesday turns the tide of his sentiments with videos of Abhishek Banerjee.

That said, Facebook clips will not shape his decision when he votes on April 29. Forwarded WhatsApp messages comprising fake clips might, especially if they involve his teen years hero Mithun Chakraborty.

On Facebook, there are layers and checks to spot synthetic content, even though there are cases where some slip through the net. It is harder to apply labels to WhatsApp groups and their encrypted chats. The platform is home to at least 530 million users, according to information shared by former Union minister for information and technology Ravi Shankar Prasad in 2021. Recent reports suggest many million additions have been made, but there is no official word on it. Synthetic content refers to digital content created or modified by AI.

In India, WhatsApp is where political content multiplies, gets a major fillip. Armed only with smartphones, political forces can create havoc. According to the Results of Comprehensive Modular Survey: Telecom, 2025 (January to March 2025), approximately 85.5 per cent of households in India possess at least one smartphone. Furthermore, around 86.3 per cent of households have access to the Internet within the household premises. Bengal is no outlier.

So while the 2014 elections were called the Twitter Elections and the 2019 elections were labelled the WhatsApp Forward Elections, the 2024 general elections saw AI technologies being integrated into old platforms. But this year’s Assembly elections across four states and one Union Territory are India’s first truly AI-powered elections.

***

“Using AI, they will try to spread lies through fake videos. We must not fall prey to such misinformation,” Prime Minister Narendra Modi recently said while addressing a rally in Bengal. His message could well be repeated verbatim by rival parties.

The AI tools that were used in 2024 have advanced a great deal. For example, back then, more than 50 million AI-generated voice calls reportedly went out in two months. These were synthetically resurrected voices. Remember how a certain Bollywood actor called your number and in his mesmerising baritone asked you to vote for a certain party?

Two years ago, a video of the late Muthuvel Karunanidhi was released in which he was seen delivering an eight-minute speech. It had a substantial reach even though the politician had died in 2018. An AI likeness of Aamir Khan was misused around the same time to mock a political party. The CPM released a video of a bearded Buddhadeb Bhattacharjee in May 2024, ahead of the general elections, when the man himself was alive but ailing. The video was titled Ei bhabeo phire asha jaay.

This time, people have been receiving robocalls from politicians in multiple languages. Voice cloning technology can be a double-edged sword — it can appeal to voters in a straightforward manner or criticise the Opposition. In the New York mayoral elections, Andrew Cuomo shared deepfake videos of his opponent Zohran Mamdani. He also used ChatGPT to generate a housing plan. Eric Adams, the incumbent mayor, had used AI to make robocalls to New Yorkers. In these he spoke Mandarin, Urdu and Yiddish. AI-generated ads were also common.

AI has been used in election campaigns in the US to target voters, write policy proposals; there is no reason to believe it is not being used in similar fashion in these parts. A quick Internet search and you will come across tender bids like this one emanating out of Tamil Nadu. The 2026 notice reads thus: “Inviting script for AI videos from media agencies… to make election awareness videos for Assembly elections.”

Divyendra Singh Jadoun has been a key figure in the Indian AI landscape. His firm, Polymath Synthetic Media Solutions, is at the cutting edge of AI technology involving the creation of digital avatars and voice cloning. In October 2020, he founded The Indian Deepfaker, which is available as a website and on social media. It gained recognition for its AI innovations in political campaigns and Bollywood.

From a one-person operation in his bedroom, his synthetic media company has grown. He has worked on campaigns for major political parties in the past, but it is unclear if he is pursuing any election-related projects at the moment.

And that’s the thing. While companies are happy to boast about their AI innovations, when it comes to elections and use of AI, they suddenly turn coy. The reason is not difficult to guess.

From what one knows of one AI-driven innovation, one can imagine the wicked genius of the rest. Jadoun’s past work offers a glimpse of how voice cloning technology can be used.

Consider a scenario where Politician X needs to address people in Hooghly, Bankura, Jalpaiguri, Murshidabad, Malda and Burdwan, either in a video or an audio clip. Earlier he would have had to record the entire message many times just to incorporate the name of each place, such as “Hooghly ke mitron...”. The politician can now make a single recording. And then sit before a camera or microphone and rattle off the entire list of places only. AI ensures that the name of each place is incorporated seamlessly into speeches addressing people of different geographies, without sounding odd.

There is no limit to what an agency can charge for its services in a campaign. “Our work has reshaped political outreach and emphasised the importance of ethical AI,” Jadoun wrote in a LinkedIn post a year ago.

***

In such a scenario, AI has come to suggest mischief. Recently, when a controversial video of Trinamool defector Humayun Kabir surfaced, he dismissed it as “AI-generated”. The authenticity of the video has not been independently verified, but since then, the AIMIM has ended its alliance with Kabir.

In this case, Kabir’s short-time ally seems to have decided it is best to be cautious. But in the tech world, there is such a thing as “liar’s dividend”. It is the phenomenon where the more we believe that what we see and hear is fake, the more we run the risk of disbelieving the authenticity of everything.

As AI generation tools keep advancing, detection tech is a step behind. Fact-checking organisations keep debunking problematic content but they take time. In the meantime, deepfakes travel at the speed of a silver bullet. Users view fake content, recognise them to be fake but do not flag them; if they did, digital platforms would be able to act.

According to reports, earlier this year, a large number of people in Japan used generative AI to help choose which candidates or parties to vote for in the general elections. An AI engineer even formed a political party called Team Mirai. Its campaign style was different — its AI Anno answered voter questions. During school exam season, it didn’t campaign near test centres and for this won the favour of parents. In some places it hosted talk-show-style rallies to drawn in GenZ voters. And it used YouTube Live to explain policies while keeping things light. Moreover, through an online political funding transparency system it developed, it disclosed income and expenditure details. Team Mirai, founded only in 2025, won 11 seats in the 2026 general elections.

But here, for now, AI in elections is good only for some backroom campaign jobs, besides stirring the misinformation pot.

In Bengal, AI algorithms are in hot water over how they have been used by the Election Commission of India to prepare SIR lists. And now, when we in Bengal are inching closer to poll dates, the EC has said AI will flag irregularities on voting days.

When asked to serve an AI wisecrack on elections, Google’s Gemini said:

Q: Why did the AI fail the Bengal Election test?

A: It tried to find a Neutral Network, but in Bengal, everyone has an opinion!

And that’s a wrap.

BOT, OF COURSE

Use of AI in elections in other countries

NEPAL: An AI-generated campaign song became hugely popular in the runup to the March 2026 elections. Balendra Shah’s supporters sang the song which means “Time’s up, Fake leaders, Game Over. We’ll ring the bell on 5 March.” According to reports, Nepalese youth used Discord, a free messaging platform mainly used by online gamers, to pick interim leader Sushila Karki in 2025. Her name was originally suggested by ChatGPT.

BANGLADESH: According to Activate Rights, Bangladesh’s 13th National Parliamentary Elections were conducted within a “digitally saturated environment where communal misinformation and AI-generated deception were weaponised to manipulate voter perception”.

UNITED STATES: In the New York mayoral elections, AI was used. While we know how Zohran Mamdani’s opponent Andrew Cuomo used AI to defame him, there was also this allegation that TikTok’s algorithm amplified pro-Mamdani content while suppressing videos backing Cuomo.

JAPAN: The political party Path to Rebirth announced in 2025 that an AI penguin would serve as its new leader, following the resignation of founder Shinji Ishimaru after poor election results. Mitsutaka Okumura, a 25-year-old graduate student at Kyoto University specialising in AI research, said, “I am the assistant leader, and the AI will act as the new leader.”

TAIWAN: It has been reported that China has deployed AI-driven sophisticated propaganda in order to shape public opinion ahead of local elections in 2026 and Presidential elections in 2028.