This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
For the second consecutive year, AI-generated images depicting singer [name withheld] at the Met Gala circulated online, causing confusion among fans and family members. The singer clarified on Instagram that she did not attend the event, noting that she was on tour. While no malicious intent was confirmed, the incident highlights the recurring issue of synthetic media being used to create misleading narratives about public figures.
Images widely reported to be AI-generated appeared to depict Katy Perry attending the Met Gala and circulated broadly online, which appeared to prompt confusion among some fans. Perry later clarified via Instagram that she had not attended the event. While no malicious actor or harm has been confirmed, this marks the second consecutive year such images have gone viral
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
The AI system was used to generate highly realistic but false images of Katy Perry at the Met Gala, misleading her mother and fans into believing she attended the event.
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 users intentionally generating and distributing deepfake images of Katy Perry using deployed AI image generation tools.
Added categories peak at 2: Minor (Epistemic Harm, Direct)
Viral AI-generated images of singer Katy Perry at the Met Gala circulated online, temporarily misleading fans and family. While demonstrating the realistic capabilities of generative AI, the incident was benign, non-malicious, and carries negligible national security implications.
The AI system acted as a tool, performing the specific function of generating images based on direct inputs without autonomous planning or context retention.
Limited Risk: The report describes AI-generated content (deepfakes) which pose a moderate risk and require transparency obligations to ensure users are informed.
The AI was used to generate highly realistic, synthetic images of a real person (Katy Perry) to depict her at an event she did not attend.
The report clearly details the creation and viral spread of the AI-generated images, the reactions of Katy Perry and her mother, and the context of her actual whereabouts. There is no ambiguity about the nature of the images or the confusion caused.
None
Technology
Highly Realistic Image Generation
AI models generated highly realistic images matching the Met Gala themes.
Lack of AI Watermarking
Generated images lacked embedded watermarks to identify them as AI-created.
Data Inputs
Abundant Public Training Data
Plentiful public images of the celebrity allowed accurate model training.
Met Gala Theme Data Availability
Contextual data about the event theme was used to make images plausible.
Human Factors
User Susceptibility to Deception
Social media users and family members easily believed realistic deepfakes.
High Trust in Visual Media
Viewers assumed photos were real without verifying the celebrity's location.
Process and Methods
Ineffective Platform Filtering
Social media platforms failed to flag or label the viral AI images.
Rapid Viral Distribution
Algorithms promoted highly engaging deepfakes quickly across networks.
Management
Prioritizing Engagement over Safety
Platforms prioritized user engagement over implementing deepfake detection.
Regulatory Environment
Lack of Deepfake Regulations
Absence of strict laws governing the creation of non-consensual deepfakes.
AI system involved
An AI system was involved, but the reports do not identify it.
Identified by automated extraction; can differ from the alleged developers above.
