This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
AI-generated deepfake images depicting Donald Trump kissing White House aide Natalie Harp went viral on social media, amassing millions of views and highlighting the growing difficulty of verifying visual content.
AI-generated images purporting to show U.S. President Donald Trump kissing White House aide Natalie Harp on a golf course spread across social media; one X post received over 10 million views. Forensic analysis and provenance checks indicated the circulating images were synthetic, and the White House called them fake. The original creator remained unidentified.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that inadvertently generate or spread incorrect or deceptive information, which can lead to inaccurate beliefs in users and undermine their autonomy. Humans that make decisions based on false beliefs can experience physical, emotional or material harms
Also classified as
The deepfake images targeted specific public figures, Donald Trump and Natalie Harp, creating a false and potentially damaging narrative about their personal relationship.
The incident involves the generation and viral spread of discrete false and misleading AI-generated images depicting a fabricated political scandal.
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 incident was caused by human actors intentionally generating and sharing deepfake images of public figures to mislead or entertain social media users post-deployment.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
An AI-generated deepfake depicting Donald Trump went viral on social media, highlighting ongoing challenges in digital content verification. The incident represents a minor national security concern, characterized as an established information-integrity threat with low coordination and no physical, sovereign, or economic disruption.
The AI systems (image generators and ChatGPT) were used as tools by human users to generate, brighten, and enhance images based on direct user inputs without autonomous planning or agency.
Limited Risk: The incident involves AI-generated content such as deepfakes, which are subject to transparency and watermarking obligations under the EU AI Act.
The primary function of the AI involved was generating a deepfake image of public figures, while ChatGPT was subsequently used as a secondary tool to brighten and enhance the image.
The report provides clear, detailed, and verified facts about the origin, spread, and detection of the deepfake images, including statements from the platforms, experts, and individuals involved in sharing the content.
None, the event is clearly a political deepfake misinformation incident.
Technology
Generative AI Image Synthesis
Generative AI tools allowed the creation of realistic fake political images.
Anatomical Generation Errors
AI image models struggle with hands, generating a hand with six fingers.
Watermark Removal Vulnerability
Blurring or degrading images could bypass standard AI detection watermarks.
Data Inputs
Low-Quality Source Images
Degraded images bypassed initial automated AI detection systems.
Human Factors
Lack of Source Verification
Users and creators shared the image without verifying its original source.
Intentional AI Image Enhancement
A user used ChatGPT to enhance and brighten the blurry fake image.
Deliberate Sharing for Engagement
Influencers used the fake image as thumbnails to attract video views.
Process and Methods
Complex Fact-Checking Process
Verifying AI content has become a complex and time-consuming manual process.
Inadequate Thumbnail Screening
Platforms failed to screen or flag AI-generated thumbnails before publication.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
