BJP Used Deepfake Videos of Manoj Tiwari to Target Haryanvi Voters in 2020 Delhi Election

In February 2020, India's BJP used purported deepfake videos of party leader Manoj Tiwari to influence voters in the Delhi legislative election. The videos reportedly showed Tiwari delivering scripted campaign messages in Haryanvi and English, which are languages he did not speak, by altering his lip movements and voice. Distributed via thousands of WhatsApp groups, the deepfakes reportedly reached millions of voters without disclosure.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.1 Disinformation, surveillance, and influence at scale

The BJP used deepfake technology to conduct a coordinated, large-scale campaign reaching 15 million people to influence political processes and voter behavior in the Delhi elections.

Additional risk subdomains

• 3.1 False or misleading information: The AI system was used to generate false videos of a politician saying things he never actually said in languages he does not speak.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The incident was caused by human political actors intentionally deploying deepfake technology post-deployment to influence voters.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The report describes the creation and distribution of deepfakes, which are explicitly classified as limited risk under the EU AI Act, requiring transparency obligations to inform users of their synthetic nature.

AI system and alleged parties

• AI system: unspecified (NVIDIA)
• AI purpose: Deepfake Video Generation; Translation
• Behaviour type: Tool
• Alleged developer: Unknown voice cloning technology developers, Unknown deepfake technology developers
• Alleged deployer: Unknown disinformation actors in India, Unknown disinformation actors, The Ideaz Factory, BJP Delhi IT Cell
• Alleged harmed parties: Manoj Tiwari, General public of India, General public, Epistemic integrity, Electoral integrity

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Severe
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Minor, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Severe
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes the creation and spread of deepfake videos used to mislead voters.

Directly caused: Two deepfake videos of politician Manoj Tiwari were created using lip-sync algorithms to make him appear to speak Haryanvi and English.

Indirectly caused: The videos were distributed across 5,800 WhatsApp groups, reaching approximately 15 million people.

Inferred additional harm: The success of this campaign may encourage other political actors to deploy similar synthetic media, potentially leading to a wider spread of malicious or deceptive content.

Democracy

Reported: The report explicitly describes concerns about the erosion of democratic norms through the use of deepfakes in election campaigns.

Directly caused: The BJP deployed fabricated videos of a candidate to influence voters in the Delhi Legislative Assembly elections, marking the debut of deepfakes in Indian election campaigns.

Indirectly caused: The campaign reached 15 million people, raising alarms among fact-checkers and policy experts about the weaponization of technology to manipulate democratic processes.

Inferred additional harm: This incident sets a precedent that could normalize the use of synthetic deception in democratic elections globally, undermining trust in democratic institutions and electoral integrity.

Epistemic

Reported: The report explicitly describes epistemic harm, noting that deepfakes make it difficult to trust what we see and hear.

Directly caused: The AI system generated realistic videos of Manoj Tiwari saying things he never actually said in languages he does not know.

Indirectly caused: The videos bypassed standard verification, with fact-checking organizations like AltNews initially unable to recognize them as doctored, thereby polluting the information ecosystem.

Inferred additional harm: The widespread use of such technology is likely to lead to a broader loss of consensus reality, where citizens can no longer distinguish between real and synthetic political communication.

People affected

• Occurrences reported: 2
• People reportedly exposed: 15000000

Potential causes

Management

• Campaign scaling prioritization: Prioritized scaling outreach efforts over potential risks of misinformation.
• Use of shadow firms: Political parties use external agencies to maintain deniability.

Technology

• Lip-sync deepfake algorithm: Allowed altering lip movements to match target audio in another language.
• Publicly available codebase: Code needed to fabricate videos is publicly available on websites.
• Lack of reliable detection tools: Tools to reliably detect deepfakes are currently unavailable to public.

Data Inputs

• Original video footage: Used real video of Manoj Tiwari to overlay synthesized mouth shapes.
• Target audio translation: Hired dubbing artist to impersonate the voice in Haryanvi language.
• Training speeches: Algorithm trained with speeches of Manoj Tiwari to learn mouth shapes.

Human Factors

• Nascent digital literacy: Low digital literacy in India makes users highly vulnerable to manipulation.
• Belief in video authenticity: Voters assumed the candidate spoke their language, creating false trust.

Process and Methods

• Distribution via WhatsApp groups: Mass distribution across 5,800 groups bypassed public scrutiny.
• Bypassing fact-checking resources: Doctored content is extremely difficult for fact-checkers to recognize.
• Lack of disclosure: No clear labeling or disclosure that the video was synthetically altered.

Regulatory Environment

• Absence of deepfake regulations: No clear legal framework in India governs or restricts political deepfakes.
• Difficulty in legal enforcement: Outlawing deepfakes is doomed to fail as they operate via shadow firms.

Information quality

• Classification confidence: High
• Reason for confidence: The report provides clear, detailed, and verified accounts of the creation, distribution, and impact of the deepfake videos, including direct quotes from both the political party (BJP) and the technical firm (The Ideaz Factory) involved.
• Ambiguities identified: The exact algorithm used (Nvidia vid2vid vs face2face) is speculated by researchers rather than definitively confirmed by the creators.
• Alternative interpretations: None. The facts of the deployment and its political intent are uncontested by the creators.

First documented deployment of deepfake technology in an Indian political campaign, where the BJP used synthetic videos of leader Manoj Tiwari to target voters. Reaching 15 million people via WhatsApp, this incident highlights the threat of automated, highly-targeted disinformation to electoral integrity and democratic sovereignty, establishing a concerning precedent for future elections.

• Overall national security impact: Substantial
• Scope: Single nation
• Primary target: India
• Alleged perpetrator: Bharatiya Janata Party

Threat characteristics

• Imminence: Long-term. The incident represents a strategic, ongoing concern regarding the manipulation of democratic processes rather than an active crisis.
• Autonomy: Human-controlled. The deepfake generation tool was directly controlled and utilized by human political actors to perform a specific function.
• Novelty: First-of-its-kind. This was the first documented real-world case of deepfakes being used in an active political election campaign to target voters.

Impact by dimension

• Physical security: Negligible. The incident involved political campaign videos and did not target physical systems, critical infrastructure, or human safety.
• Information security: Substantial. Represented a substantial information operations campaign using deepfakes to target voters, challenging fact-checkers and polluting the information ecosystem at scale.
• Sovereignty: Substantial. The intentional deployment of synthetic media to influence a democratic election represents a notable impact on core government processes and electoral integrity.
• Economic security: Negligible. No direct threats to financial stability, strategic industries, or technological competitive advantage were reported.
• Societal stability: Minor. While raising concerns about democratic norms and trust, the incident did not cause physical harm, widespread civil unrest, or systematic human rights violations.