This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Linked social media accounts under the name 'Citizens Against Mamdani' utilized generative AI, including OpenAI's Sora, to create and disseminate deepfake videos of fictional New Yorkers criticizing mayor-elect Zohran Mamdani. The campaign aimed to simulate grassroots opposition through automated astroturfing. Cybersecurity firms confirmed the high likelihood of the videos being AI-generated, highlighting the use of LLMs and text-to-video tools as a force multiplier for disinformation.
In New York, linked "Citizens Against Mamdani" accounts reportedly posted AI-generated videos of fictional constituents criticizing then-mayor-elect Zohran Mamdani across Instagram, TikTok, and X. Some videos reportedly displayed a visible Sora watermark, and forensic reviewers reportedly assessed at least one clip as highly likely to be a deepfake. The campaign reportedly appeared to simulate grassroots opposition and drew substantial engagement.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to conduct large-scale disinformation campaigns, malicious surveillance, or targeted and sophisticated automated censorship and propaganda, with the aim of manipulating political processes, public opinion, and behavior.
Also classified as
The automated astroturfing creates a 'mirage of political popularity' that pollutes the social media information ecosystem.
The incident involves a coordinated campaign using AI-generated deepfakes to simulate grassroots political opposition and manipulate public opinion.
Human
Due to a decision or action made by humans
Intentional
Due to an expected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed
The campaign was intentionally launched by human actors using deployed AI models to generate political disinformation.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
A coordinated digital influence campaign utilized generative AI tools, including OpenAI's Sora, to create deepfake videos of fictional New Yorkers criticizing NYC mayor-elect Zohran Mamdani. The incident represents an evolved capability in automated political astroturfing, though its national security impact remains minor and localized.
The report describes the AI as a 'force multiplier' where users 'define an LLM with a sentiment' and use 'text-to-video to create hyperrealistic videos'. The AI systems operated as tools executing specific generation functions based on direct human prompts.
Risk Level 3. Limited Risk: The incident involves AI-generated content such as deepfakes. The report notes the presence of a 'Sora watermark' on some videos, which aligns with transparency obligations for synthetic media.
The AI system Sora was primarily used to generate deepfake videos of fictional citizens, which were then posted to social media platforms to simulate political opposition.
The report clearly details the nature of the campaign, the platforms used, the specific AI technologies involved (Sora, LLMs), and the assessments by cybersecurity firms (GetReal Security, Reality Defender) confirming the synthetic nature of the media.
None. The evidence of AI generation (Sora watermark, 99% deepfake score) is overwhelming.
Technology
Advanced Text-to-Video Tools
Generative AI like Sora makes hyperrealistic video creation cheap and easy.
LLM Script Generation
LLMs allow rapid generation of diverse and convincing scripts at scale.
Imperfect Detection Tools
No surefire way to confirm content is AI-generated without advanced tech.
Data Inputs
Synthetic Persona Data
Algorithms use realistic accents and faces to mimic diverse populations.
Unlabeled AI Content
Missing or easily ignored watermarks make identifying deepfakes hard.
Human Factors
User Susceptibility to Deepfakes
Online users take fake videos seriously, believing the illusion of support.
Low Barrier to Malicious Use
Only a few prompts are needed to generate political disinformation campaigns.
Process and Methods
Automated Astroturfing
Combining LLMs and deepfakes automates the creation of fake support online.
Inadequate Platform Moderation
Social media platforms host deepfakes without immediate detection or removal.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
