Deepfakes, Dodgy Ads and Foreign Bots: The AI Test Facing Indian Democracy

Navyug News | Technology & Politics | October 9, 2026

In 2024, India held the largest election the world has ever seen, and it was also one of the first at that scale where cheap AI tools were available to almost anyone with a smartphone. The result was a campaign season full of synthetic voices, resurrected leaders, doctored celebrity clips and AI-made advertisements. Two years on, the lessons are still being absorbed by regulators, platforms and voters. This article sticks to what has actually been documented.

AI is already a standard campaign tool

Not every use of AI in politics is sinister. Parties use it to translate speeches into regional languages, produce slogans, and make videos faster and cheaper. Some uses are openly theatrical. Ahead of the early-2025 Delhi assembly election, the Aam Aadmi Party shared an AI-generated video showing B.R. Ambedkar blessing Arvind Kejriwal, and the BJP accused the party of misusing technology to mislead and mock. Before the 2024 general election, AI was also used to bring back deceased politicians, including the Tamil Nadu leader J. Jayalalithaa, in videos meant to sway living voters, according to the AI Incident Database.

The line between harmless gimmick and deception is where the debate begins.

Deepfakes of celebrities and leaders

The most widely reported deception cases involved fabricated endorsements. In April 2024, AI-generated videos showing the Bollywood actors Aamir Khan and Ranveer Singh criticising Prime Minister Narendra Modi and urging people to vote for the Congress went viral. Google's own policy announcement later that year cited these videos as an example of what worries regulators.

Fact-checkers raise a second danger that gets less attention. Kritika Goel of Logically Facts has pointed to the "liar's dividend": once convincing fakes are common, a politician caught on genuine audio can simply claim it was manufactured.

The Meta ads experiment

The clearest evidence about platform weaknesses came from a test run by India Civil Watch International and the corporate accountability group Ekō. During phases three and four of the 2024 election, they submitted 22 ads to Meta, each paired with an image made using tools such as Stable Diffusion, Midjourney and DALL-E. The ads carried inflammatory content, including hate narratives aimed at Muslims and a call for an opposition leader's execution, and were aimed at 189 constituencies during the 48-hour silence period when election advertising is barred.

Meta approved 14 of the 22 within 24 hours, and, according to the researchers, none was flagged as AI-generated. The investigators pulled the ads before publication, so no real user saw them. Meta did not comment on the report itself, but said it requires political advertisers to disclose AI use and that it had taken steps for the Indian election.

The same research network has documented a separate problem: roughly two dozen "shadow advertisers" that spent about one million dollars in the first three months of 2024 promoting memes and short videos that dehumanised minorities and opposition parties. Some of these advertisers were described as aligned with political parties, particularly the BJP, a characterisation that the report's authors, not this publication, make.

Foreign influence: what the record shows

Foreign interference is a legitimate worry, and there is one well-documented case. In a report published on May 30, 2024, OpenAI said it had disrupted an operation run by STOIC, a political campaign management firm in Israel, nicknamed "Zero Zeno". The network used OpenAI's models to produce articles and comments for Instagram, Facebook, X and websites. In early May it began targeting Indian audiences in English, with comments critical of the ruling BJP and favourable to the Congress. OpenAI said the effort drew little genuine engagement and that it disrupted the activity within about a day of the India-focused push. Meta reported separately that it had removed a STOIC-linked network of 510 Facebook accounts and 32 Instagram accounts.

At the time, Minister of State for IT Rajeev Chandrasekhar responded by saying the BJP was a target of foreign influence operations carried out by or on behalf of some Indian political parties. The report itself made no claim about Indian parties, and that accusation should be read as the minister's political claim.

Two points are worth underlining. First, this is a commercial firm abroad using AI tools, not a government or an AI company acting in India. Second, in this case it was a US AI company and a US social media company that found and shut the network down. Anyone alleging otherwise about a platform needs evidence of the same quality.

What platforms promise, and where they fall short

Platform rules have tightened since 2023. Meta requires disclosure when a political or social-issue ad contains a photorealistic image, video or realistic audio that has been digitally altered to show a real person doing or saying something they did not. Google introduced a similar checkbox disclosure for election ads containing altered or synthetic content. Google also restricted its Gemini chatbot from answering election-related questions.

Enforcement is the weak point. Reporting on Hokage Modi Sama's AI-made political ads found that Meta did not treat hashtags like #aiartwork as acceptable disclosure and expects its own "Digitally created" label instead. The Ekō test showed that the detection tools were not catching AI images at the point of ad approval.

What Indian rules now require

India has layered several measures on top of each other:

• May 6, 2024: The Election Commission of India (ECI) issued guidelines asking parties to avoid deepfakes and distorted AI content.

• January 2025: The ECI advised parties, candidates and star campaigners to label AI-generated or synthetic content prominently.

• October 24, 2025: A further ECI advisory asked for labels such as "AI-Generated" or "Synthetic Content" covering at least 10 percent of the visible area of an image or video (or the first 10 percent of an audio clip), and for parties to keep records of AI material they publish.

• February 2026: The IT Ministry notified amendments to the IT Rules, in force from February 20, 2026, bringing "synthetically generated information" into the due-diligence duties of platforms. These cover mandatory labelling and provenance metadata, with much shorter takedown windows, including three hours in some categories. Platforms that fail to comply risk losing safe-harbour protection under Section 79 of the IT Act.

Two gaps stand out. The ECI measures are advisories, and the sources reviewed for this article do not show how often they have been enforced. And the IT Rules definition covers audio and visual content only, so AI-written text, such as the comment-farming in the Zero Zeno case, falls outside it.

What readers can do

• Look for an AI label, but do not treat its absence as proof that content is real.

• Be suspicious of any video of a famous person endorsing a party, especially if it is short, shocking and shared widely on messaging apps.

• Check suspicious clips with established fact-checkers before forwarding.

• Remember that a screenshot with a name on it is not evidence. Look for the original source.

  
  

The bottom line

The documented problems in Indian politics are real: convincing fake endorsements, platform ad systems that missed AI images, and at least one foreign commercial influence operation. They are also specific, dated and traceable. That is how a healthy public debate should work. Claims about who is behind political AI content should rest on reports like these, not on rumours that spread faster than any deepfake.