Hero image: Lara Jameson / Pexels
AI Deepfakes Fuel Disinformation in Maharashtra Protests
Independent reporting from the OECD AI Policy Observatory and other sources reveals how AI-generated deepfakes are being weaponized during protests in Maharashtra, blurring the line between fact and fiction and complicating crisis response. The evidence points to coordinated disinformation campaigns that exploit social platforms, erode public trust, and challenge institutional responses.
The claim that AI-generated deepfakes are being used to spread disinformation during protests in Maharashtra is not speculative—it is documented. The OECD AI Policy Observatory has published a detailed report examining how synthetic media is being deployed in real time during the CJP protests, raising urgent questions about detection, accountability, and governance. This synthesis examines what is known, how it is being framed across independent outlets, and what the combined evidence reveals about the evolving threat landscape. Where reporting converges or diverges, it is highlighted; where gaps remain, they are acknowledged.
—
Context: The Rise of AI-Generated Deepfakes in Political Protests
The use of AI-generated deepfakes to manipulate public perception during political events is not new, but its scale and sophistication have accelerated with advances in generative AI models. Synthetic media—including audio, video, and images—can now be produced with minimal technical expertise, enabling rapid, low-cost disinformation campaigns. These tools are particularly effective during protests, where emotions run high and verification is often delayed.
Historically, deepfakes were confined to niche forums and required significant computational resources. Today, open-source tools and cloud-based AI services have democratized access, allowing actors with limited technical skills to generate convincing fakes that mimic public figures, activists, or even bystanders. The result is a flood of synthetic content that can be weaponized to inflame tensions, discredit organizers, or mislead authorities.
—
What the OECD AI Policy Observatory Reports About Maharashtra
The OECD AI Policy Observatory’s report, AI-Generated Deepfakes Fuel Disinformation During CJP Protests in Maharashtra, provides the most detailed account to date of how synthetic media is being used in the current crisis. According to the Observatory, deepfakes are being circulated on social media platforms to falsely attribute incendiary statements to protest leaders, simulate violent confrontations, and fabricate evidence of police brutality. These clips are often designed to go viral, exploiting algorithmic amplification to reach millions within hours.
The report highlights a specific incident in which a deepfake video purportedly showed a prominent activist inciting violence during a protest in Mumbai. The video, which was later debunked by fact-checkers, was shared by thousands of accounts within minutes, including several with known links to coordinated inauthentic behavior. The Observatory notes that the video’s rapid spread was facilitated by platform algorithms that prioritize engagement over veracity, particularly during breaking news events.
Additionally, the Observatory identifies a pattern of cross-platform amplification, where deepfakes are first posted on Twitter/X, then reposted on Facebook and WhatsApp, and finally amplified by regional news outlets that lack robust verification processes. This multi-stage dissemination strategy increases the likelihood that the disinformation will be treated as credible, even after debunking efforts begin.
—
Cross-Outlet Comparison: How Independent Sources Frame AI Deepfake Disinformation
While the OECD AI Policy Observatory provides the most granular technical and procedural analysis, other independent outlets have framed the issue through different lenses. For instance, The Hindu emphasized the human impact, reporting on how local communities in Maharashtra are struggling to distinguish between real and synthetic content, particularly in areas with limited digital literacy. Their coverage highlights the psychological toll of disinformation, including fear, mistrust, and social fragmentation.
In contrast, Scroll.in focused on the institutional response, noting that Maharashtra’s police and election authorities have begun deploying AI-based detection tools to identify deepfakes in real time. However, their reporting also underscores the limitations of these tools, which often lag behind the latest generative models and can produce false positives under pressure.
Meanwhile, IndiaSpend took a data-driven approach, analyzing the network patterns behind deepfake dissemination. Their investigation found that a small number of highly active accounts—many of which exhibit bot-like behavior—are responsible for a disproportionate share of deepfake sharing. This suggests that the disinformation ecosystem is not entirely organic but is being amplified by coordinated actors.
Taken together, these reports paint a picture of a crisis that is both technological and social: AI tools are enabling new forms of manipulation, but the vulnerabilities lie as much in human cognition and institutional capacity as in the technology itself.
—
Divergences in Emphasis
One area of divergence is the role of foreign actors. While The Hindu and Scroll.in focus primarily on domestic dynamics—including political polarization and institutional preparedness—IndiaSpend’s analysis does not rule out external influence but does not provide conclusive evidence. The OECD report remains agnostic on attribution, noting only that the deepfakes are being amplified by both domestic and cross-border networks.
Another difference lies in the assessment of platform responsibility. Scroll.in is critical of social media companies for failing to act swiftly, while The Hindu takes a more nuanced view, acknowledging the challenges platforms face in moderating content in real time. The OECD report does not directly address platform culpability but highlights the need for better detection and labeling mechanisms.
—
The Mechanism: How AI Deepfakes Are Deployed During Protests
Production and Customization
According to the OECD AI Policy Observatory, the production of deepfakes during the Maharashtra protests follows a predictable workflow. First, actors identify a target—often a public figure, a protest leader, or a law enforcement officer—and gather existing footage or audio of them. This source material is then used to train a generative model, which can produce new content that mimics the target’s voice, facial expressions, and mannerisms.
The Observatory notes that many of these models are fine-tuned versions of open-source tools, such as Stable Diffusion for images or VITS for audio, which have been adapted for regional languages and dialects. This customization increases the realism of the fakes and makes them more likely to be believed by local audiences.
Distribution and Amplification
Once generated, deepfakes are uploaded to social media platforms with misleading captions designed to trigger emotional responses. For example, a deepfake video of a police officer allegedly assaulting a protester might be captioned “Police brutality exposed!” or “See how they treat our children!” These captions are optimized for virality, using emotionally charged language and hashtags that align with protest narratives.
IndiaSpend’s analysis shows that these posts are then amplified by a mix of automated accounts and real users who share them without verification. The accounts often exhibit bot-like behavior, including rapid posting, repetitive messaging, and coordination with other suspicious accounts. This creates the illusion of widespread outrage, which in turn drives further engagement and visibility.
Feedback Loops and Escalation
The OECD report describes a feedback loop in which deepfakes not only spread disinformation but also provoke real-world reactions that are then captured and recontextualized. For instance, a deepfake video might spark a protest, which is then recorded and shared as “evidence” of the original claim. This creates a cycle in which synthetic and real content become indistinguishable, further eroding trust.
—
Who Is Affected: From Local Communities to National Institutions
Local Communities and Grassroots Activists
The Hindu reports that local communities in Maharashtra are among the hardest hit by deepfake disinformation. In rural and semi-urban areas, where digital literacy is lower and access to fact-checking resources is limited, false narratives can take root quickly. Residents interviewed by The Hindu described feeling overwhelmed by the volume of conflicting information, leading to distrust not only in the government but also in fellow citizens.
Grassroots activists, particularly those associated with the CJP protests, have become primary targets. Deepfakes are used to fabricate evidence of illegal activity, hate speech, or financial impropriety, which can lead to harassment, legal threats, or even physical violence. The Observatory notes that some activists have received death threats based on deepfake content, forcing them to go offline or relocate.
Law Enforcement and Emergency Responders
Scroll.in highlights the strain on Maharashtra’s police and emergency services, who must respond to both real and fabricated incidents. During protests, officers are often forced to prioritize debunking deepfakes over addressing genuine public safety concerns. This not only diverts resources but also risks undermining public confidence in law enforcement.
The Observatory adds that deepfakes are also being used to impersonate police officers, issuing false orders or making inflammatory statements that escalate tensions. In one documented case, a deepfake audio clip purportedly showed a senior officer instructing officers to use excessive force. The clip was widely shared before being debunked, but not before it had already fueled further unrest.
National Institutions and Democratic Processes
While the immediate impact is localized, the disinformation campaign has national implications. IndiaSpend warns that the tactics used in Maharashtra—coordinated amplification, emotional manipulation, and cross-platform dissemination—could be replicated in other states ahead of elections. The report suggests that if unchecked, deepfake-driven disinformation could distort public discourse, influence voting behavior, and erode democratic norms.
The OECD report echoes this concern, noting that the proliferation of synthetic media threatens to destabilize trust in institutions, media, and electoral processes. It calls for a coordinated response that goes beyond platform-level fixes to address the underlying drivers of disinformation, including algorithmic amplification and societal polarization.
—
How Disinformation Spreads: Platforms, Networks, and Algorithms
Platform Incentives and Amplification
The OECD AI Policy Observatory identifies platform algorithms as a key enabler of deepfake spread. During breaking news events like protests, algorithms prioritize content that generates high engagement, regardless of its veracity. This creates a perverse incentive: the more inflammatory or sensational the content, the more likely it is to be promoted.
Scroll.in corroborates this, noting that even after deepfakes are flagged or debunked, the original posts often remain visible and continue to circulate in private groups or through reshares. The platforms’ reluctance to remove content—citing concerns about free expression—further delays corrective action.
Networked Amplification
IndiaSpend’s network analysis reveals that deepfakes are not spread evenly across platforms but are concentrated within specific communities. These communities are often insulated from counter-speech, creating echo chambers where disinformation can fester. The report identifies several “super-spreader” accounts—real users with large followings who repeatedly share deepfakes without context or verification.
The Observatory adds that these accounts frequently operate across multiple platforms, using cross-posting and link-sharing to maximize reach. For example, a deepfake video posted on Twitter/X might be embedded in a Facebook post, shared in a WhatsApp group, and then reposted on YouTube with a new title designed to mislead search algorithms.
Cross-Platform Coordination
The combined evidence suggests that deepfake disinformation is not a platform-specific problem but a systemic one. The Hindu reports that regional news outlets, particularly those with partisan leanings, have amplified deepfakes by treating them as legitimate news. In some cases, these outlets have published deepfakes without verification, citing social media as a source. This blurs the line between user-generated content and professional journalism, further confusing audiences.
The OECD report warns that this cross-platform ecosystem makes it nearly impossible for any single platform or institution to contain the spread of disinformation. It calls for a collaborative approach that includes platforms, governments, civil society, and the public.
—
Red Flags and Debunking Checklist for AI-Generated Media
Distinguishing AI-generated deepfakes from authentic media requires a combination of technical scrutiny and contextual awareness. Below is a checklist of red flags and verification steps, synthesized from the OECD report and independent fact-checkers.
- Unnatural Facial Movements: Look for inconsistencies in blinking, eye movement, or facial expressions. AI-generated faces often exhibit subtle artifacts, such as asymmetrical expressions or unnatural lip sync.
- Audio Anomalies: Listen for robotic or monotone speech patterns, unnatural pauses, or background noise that doesn’t match the claimed location. Tools like Resemble AI or ElevenLabs detectors can help identify synthetic audio.
- Inconsistent Lighting and Shadows: AI-generated images or videos may have unnatural lighting, such as shadows that don’t align with the claimed time of day or location.
- Background Artifacts: Look for blurring, pixelation, or unnatural textures in the background, particularly around edges or fine details.
- Source Verification: Reverse-image search the video or image to see if it has been previously debunked. Check the account’s history for signs of bot-like behavior, such as rapid posting or repetitive messaging.
- Contextual Inconsistencies: Does the content align with known facts about the event or individual? For example, if a video claims to show a protest in Mumbai but includes landmarks from a different city, it is likely fake.
- Emotional Manipulation: Be wary of content designed to provoke strong emotions, such as outrage or fear. Disinformation often relies on emotional triggers to bypass critical thinking.
- Platform Signals: Check if the content has been labeled or debunked by platform fact-checkers or independent organizations. Look for warnings or context boxes added by the platform.
- Metadata Scrutiny: Use tools like InVID or FotoForensics to analyze metadata, which may reveal inconsistencies in timestamps, locations, or device information.
- Cross-Platform Verification: If the content appears on multiple platforms, compare the versions. Discrepancies in captions, timestamps, or visual details may indicate manipulation.
If you encounter content that exhibits multiple red flags, treat it as unverified until confirmed by multiple trusted sources. Do not share or amplify it without context.
—
Institutional Responses: Governments, Tech Platforms, and Civil Society
Government Actions
Scroll.in reports that the Maharashtra government has launched a task force to monitor and counter deepfake disinformation during the CJP protests. The task force includes representatives from the police, cybersecurity agencies, and civil society organizations. Its mandate includes identifying deepfakes in real time, issuing public advisories, and coordinating with social media platforms to remove harmful content.
However, The Hindu notes that the task force’s efforts are hampered by limited resources and legal ambiguities. For example, authorities struggle to obtain user data from platforms like WhatsApp, which encrypts messages end-to-end. This makes it difficult to trace the origin of deepfakes or hold perpetrators accountable.
The OECD report adds that while some states have introduced legislation to criminalize deepfake creation and distribution, enforcement remains inconsistent. The report calls for a national framework that clarifies jurisdiction, streamlines reporting mechanisms, and provides legal protections for whistleblowers and journalists.
Tech Platform Responses
IndiaSpend finds that social media platforms have taken uneven steps to address deepfake disinformation. Twitter/X has introduced warning labels for manipulated media, while Facebook and WhatsApp have partnered with fact-checking organizations to debunk false claims. However, these efforts are often reactive, relying on user reports or third-party flagging rather than proactive detection.
The OECD report highlights a critical gap: platforms lack standardized tools to detect deepfakes in regional languages and dialects. Most detection systems are trained on English-language datasets, making them ineffective for content in Marathi or other Indian languages. The report urges platforms to invest in localized AI models and collaborate with regional experts to improve detection accuracy.
Civil Society and Media Initiatives
The Hindu profiles several grassroots organizations that are working to counter deepfake disinformation. These groups conduct digital literacy workshops in rural areas, train journalists in verification techniques, and publish fact-checks in regional languages. Their work is vital in communities where institutional responses are slow or absent.
Scroll.in notes that some media outlets have adopted “pre-bunking” strategies, publishing explainers on deepfake tactics before they become widespread. For example, newsrooms are running articles on how to spot AI-generated content, using real examples from the protests to illustrate warning signs.
The OECD report emphasizes the need for a multi-stakeholder approach, where governments, platforms, and civil society work together to build resilience against disinformation. It warns that piecemeal responses will fail unless they address the root causes of vulnerability, including algorithmic amplification and societal polarization.
—
Original Analysis: What the Combined Evidence Reveals About the Threat Landscape
Taken together, the reports from the OECD AI Policy Observatory, The Hindu, Scroll.in, and IndiaSpend reveal a threat landscape that is both rapidly evolving and deeply entrenched. The evidence suggests that AI deepfakes are not merely a tool of disinformation but a symptom of broader systemic vulnerabilities—platform incentives that reward engagement over truth, institutional responses that are reactive rather than preventive, and societal divisions that make audiences susceptible to manipulation.
One of the most striking patterns is the convergence of production, distribution, and amplification into a single, self-reinforcing ecosystem. Deepfakes are not created in isolation; they are designed to exploit platform algorithms, networked communities, and emotional triggers. This makes them particularly effective during protests, where real-time verification is difficult and public emotions are already heightened.
Another key insight is the role of regionalization. Unlike earlier waves of disinformation, which were often global in scope, the Maharashtra deepfake campaign is highly localized, tailored to Marathi-language audiences and regional political dynamics. This localization increases the impact of the disinformation while making it harder for national or international actors to detect and counter.
The reports also highlight a critical paradox: the same tools that enable deepfake creation—open-source AI models, cloud computing, and social media platforms—are also being used to combat disinformation. Fact-checkers, digital literacy campaigns, and AI-based detection tools are all part of a growing ecosystem of resistance. However, these efforts are outpaced by the speed and scale of the disinformation campaigns, suggesting that technological solutions alone will not suffice.
Finally, the evidence points to a governance gap. While some states have introduced legislation and task forces, enforcement is inconsistent, and legal frameworks are often outdated. The result is a patchwork of responses that leaves gaps for bad actors to exploit. A national strategy—one that clarifies jurisdiction, standardizes detection tools, and protects free expression while curbing harm—is urgently needed.
In short, the Maharashtra protests are not an anomaly but a harbinger. They demonstrate how AI-generated deepfakes can be weaponized in real time to distort public discourse, undermine trust, and escalate conflict. The question is no longer whether this will happen again, but how prepared institutions, platforms, and citizens are to respond.
—
What Can Be Done: Policy, Technology, and Public Action
Policy and Regulation
The OECD report calls for a tiered regulatory approach that balances innovation with accountability. At the national level, it recommends the creation of a dedicated agency to monitor and respond to AI-driven disinformation, with the power to issue binding guidelines for platforms and enforce penalties for non-compliance. The agency would also be responsible for coordinating with state governments, law enforcement, and civil society to ensure a unified response.
Scroll.in adds that state-level task forces should be given greater authority to investigate and prosecute deepfake creators, including the ability to compel platform cooperation. It also urges the government to invest in public awareness campaigns that educate citizens about the risks of deepfakes and how to spot them.
The report warns against over-reliance on criminalization, noting that punitive measures alone will not address the root causes of disinformation. Instead, it advocates for a combination of carrots and sticks—rewarding platforms and creators who adopt best practices while penalizing those who enable harm.
Technological Solutions
Platforms must prioritize the development of localized detection tools that can identify deepfakes in regional languages and dialects. The OECD report highlights the need for open-source detection models that can be deployed by fact-checkers, journalists, and civil society organizations. These tools should be integrated into platform algorithms to flag suspicious content in real time.
IndiaSpend suggests that platforms should also implement “pre-emptive labeling,” where content is flagged as potentially manipulated before it goes viral. This would reduce the reach of deepfakes while giving users time to verify the information. Additionally, platforms should expand their fact-checking partnerships to include regional organizations that can provide context and debunking in local languages.
The report also calls for greater transparency in platform algorithms, particularly around how content is amplified during breaking news events. Users should be able to see why a piece of content is being recommended and have the ability to opt out of algorithmic amplification.
Public Action and Media Literacy
The Hindu emphasizes the role of digital literacy in building resilience against deepfake disinformation. It recommends that schools and community centers incorporate media literacy into their curricula, teaching students how to evaluate sources, spot manipulation, and verify information. Workshops should be conducted in regional languages and tailored to different age groups.
The OECD report adds that civil society organizations should lead grassroots campaigns to counter disinformation, particularly in rural and semi-urban areas. These campaigns should focus on building trust in local institutions, such as schools, religious organizations, and community leaders, who can serve as trusted sources of information.
Finally, the public must adopt a culture of skepticism and verification. This means pausing before sharing content, checking multiple sources, and seeking out context before forming an opinion. It also means holding institutions and platforms accountable when they fail to act.
—
FAQ
How can I tell if a video or image is a deepfake?
Look for unnatural facial movements, inconsistent lighting, or audio anomalies such as robotic speech. Use reverse-image search tools and metadata analysis to verify the content’s origin. Be wary of content designed to provoke strong emotions, as this is a common tactic in disinformation campaigns.
Are deepfakes illegal in India?
Some states have introduced legislation to criminalize deepfake creation and distribution, but enforcement is inconsistent. At the national level, there is no specific law targeting deepfakes, though existing laws on defamation, impersonation, and cybercrime may apply. A national framework is urgently needed to clarify jurisdiction and penalties.
Can social media platforms stop deepfakes from spreading?
Platforms can reduce the spread of deepfakes by implementing proactive detection tools, pre-emptive labeling, and algorithmic adjustments that prioritize veracity over engagement. However, no single platform can solve the problem alone. A collaborative approach involving governments, civil society, and the public is essential.
What role do regional languages play in deepfake disinformation?
Regional languages are a key enabler of deepfake disinformation in India. Most detection tools are trained on English-language datasets, making them ineffective for content in Marathi, Hindi, or other regional languages. This localization increases the impact of the disinformation while making it harder to detect and counter.
What can I do if I encounter a deepfake?
Do not share or amplify the content without verification. Report it to the platform and flag it for fact-checking organizations. If it involves threats or harassment, contact local authorities. Educate others about the red flags and encourage a culture of skepticism and verification.
—