Incident Overview
On November 5, 2025, a coordinated campaign involving anonymous social media accounts under the name 'Citizens Against Mamdani' posted AI-generated videos depicting fictional New York residents criticizing Zohran Mamdani, then the mayor-elect of New York City. These accounts aimed to simulate grassroots opposition to Mamdani, creating the illusion of widespread public dissent. The campaign drew significant engagement, amplifying its reach and influence.
The videos featured synthetic personas designed to appear as ordinary citizens expressing dissatisfaction with Mamdani’s policies or character. Some clips displayed a visible watermark from OpenAI’s Sora, a text-to-video AI model, indicating the use of advanced generative AI tools. Forensic analysis confirmed that at least one video was highly likely to be a deepfake, further validating concerns about the authenticity of the content.
This incident marked an early and prominent example of AI-generated 'vox pop' content—short video clips of fabricated individuals—being deployed in a U.S. political context. It predated similar operations, such as the AI-generated videos used in the Kosovo election interference case, demonstrating the growing sophistication of disinformation tactics in electoral settings.
The use of AI-generated content also raised concerns about the erosion of democratic integrity, as synthetic voices manipulated perceptions of political support.
The 'Citizens Against Mamdani' accounts operated anonymously, obscuring the identities of the operators behind the campaign. While the specific actors responsible remain unidentified, the incident implicates unknown social media account operators and potential information manipulation actors. The deployment of AI tools like Sora and synthetic audio generation technologies suggests a deliberate strategy to exploit generative AI capabilities for political influence.
The incident was first reported by media outlets and documented in the AI Incident Database (AIID) under Incident ID 1425. The timeline was established based on a recovered Instagram post from the citizensagainstmamdani account, which included a visible Sora watermark and an on-platform posting date of November 5, 2025. This date represents the earliest verified public posting linked to the campaign.
The use of deepfake technology and synthetic media in this context aligns with broader trends in AI-driven disinformation, where fabricated content is weaponized to distort public perception and influence electoral outcomes.
Possible Root Cause
The primary root cause of this incident lies in the misuse of generative AI technologies, specifically text-to-video models like OpenAI’s Sora and synthetic audio generation tools. These systems were designed for creative and legitimate purposes but were repurposed to produce deceptive content. The visible Sora watermark on some videos indicates that the operators either used publicly available AI tools or accessed versions of the technology that retained identifiable markers, though this did not prevent the content from being perceived as authentic by viewers.
The campaign’s success in simulating grassroots opposition relied on the plausibility of the AI-generated personas. The videos depicted fictional New Yorkers with fabricated grievances, leveraging the emotional and psychological impact of visual and auditory deepfakes. The use of local accents, settings, and references to Mamdani’s policies enhanced the credibility of the content, making it difficult for casual observers to distinguish it from genuine citizen feedback.
The anonymity of the 'Citizens Against Mamdani' accounts facilitated the operation without direct attribution, allowing the operators to avoid accountability. This lack of transparency is a common tactic in influence operations, enabling actors to disseminate misleading content while remaining untraceable. The coordination across multiple platforms—Instagram, TikTok, and X—further amplified the campaign’s reach, as each platform’s algorithmic systems promoted the content to diverse audiences.
The incident also highlights the role of algorithmic amplification in spreading disinformation.
While the specific motivations of the operators remain unclear, the campaign aligns with broader trends in political disinformation, where synthetic media is used to manipulate public opinion. The targeting of Mamdani suggests a deliberate attempt to undermine his electoral legitimacy or influence voter perceptions ahead of or during his tenure. The use of AI-generated vox pops as a precursor to more overt interference tactics indicates a strategic escalation in the tactics employed by information manipulation actors.
Takeaways
The use of generative AI to create synthetic personas and fabricated grievances demonstrates how easily deepfakes can be deployed to distort public discourse.
The visibility of the Sora watermark in some videos presents a mixed outcome.
The anonymity of the operators behind the 'Citizens Against Mamdani' accounts highlights the challenges of attributing responsibility in AI-driven disinformation campaigns. Legal frameworks may need to evolve to address the unique challenges posed by AI-generated content, including clearer definitions of harm and accountability for synthetic media.
The incident serves as a case study in the broader risks associated with AI-generated misinformation, particularly in electoral settings. It demonstrates how synthetic media can be weaponized to create artificial narratives, manipulate voter perceptions, and undermine political candidates. As AI capabilities continue to advance, the potential for such tactics to scale and evolve poses significant threats to democratic integrity worldwide.
For policymakers, tech companies, and civil society, this incident reinforces the importance of proactive measures to counter AI-driven disinformation. The 'Citizens Against Mamdani' case illustrates that the battle against AI-generated influence operations requires a multifaceted approach, combining technological innovation, regulatory action, and public awareness.
While generative AI tools like Sora offer transformative potential for creativity and communication, their dual-use nature demands careful consideration of how they are deployed.
Sources
Incident summary from the AI Incident Database, used under CC BY-SA 4.0.