This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Following a real-world jewel heist at the Louvre on October 19, 2025, fabricated videos depicting the robbery circulated widely on social media platforms including Facebook, Douyin, and RedNote. Digital forensics analysis confirmed the videos were AI-generated, citing morphing artifacts and the presence of OpenAI's Sora watermarks. The incident highlights the use of generative AI tools to create and disseminate misinformation following high-profile news events.
After the October 19, 2025 jewel heist at the Louvre, purported deepfakes were reported to have circulated widely online, allegedly depicting the robbery despite being fabricated.. The clips were reportedly posted on Facebook, Douyin and RedNote, and they were noted to have featured morphing artifacts, disappearing objects, and partial Sora watermarks. AI Forensics and independent checks reportedly found the scenes inconsistent with the real Apollo Gallery, confirming the footage was not real.
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 incident involved the creation and viral dissemination of AI-generated fake videos depicting a real-world museum heist, leading to public misinformation.
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 a human actor intentionally using OpenAI's Sora video generator to create and distribute fake footage of a real-world heist.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
AI-generated video depicting a real-world heist at the Louvre circulated on social media, demonstrating how advanced text-to-video tools like OpenAI's Sora can be rapidly deployed to create realistic misinformation following major news events. While the incident posed negligible threats to critical infrastructure or sovereignty, it highlights an evolving capability in epistemic harm and public trust erosion.
The AI system, OpenAI's Sora, acted as a tool to generate video clips based on user inputs. The report notes it is an 'AI video-generation tool' used to produce the fabricated heist footage.
Limited Risk: The system is an AI video-generation tool used to create deepfakes or AI-generated content. Under the EU AI Act, providers of such systems must ensure that outputs are marked as AI-generated, which aligns with the 'Sora watermarks' mentioned in the report.
The AI system was used to generate highly realistic but entirely fabricated video clips of a real-world heist to be shared on social media.
The report clearly identifies the AI system involved (OpenAI's Sora) via watermarks and forensic analysis, details the platforms where the content circulated, and provides specific visual discrepancies that proved the videos were fake. There is little ambiguity regarding the events or the AI's role.
None
Technology
Sora Video Generation
OpenAI's Sora was used to generate highly realistic fake heist videos.
Obscurable Watermarks
Sora's watermarks were partially obscured, making identification harder.
AI Generation Artifacts
Morphed hands and disappearing objects revealed the videos were AI-made.
Human Factors
Deceptive Content Sharing
A user shared fake footage to a large following to gain engagement.
Lack of Public Verification
Audiences widely shared the video without verifying its authenticity.
Process and Methods
Inadequate Content Moderation
Social media platforms failed to detect and filter the fake heist videos.
Weak Watermark Detection
Platforms did not scan for or flag the obscured OpenAI watermarks.
Management
Prioritizing Engagement
Platforms prioritized viral reach over immediate content verification.
Regulatory Environment
Lack of AI Content Mandates
No strict regulations exist to enforce labeling of AI-generated media.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
