This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
Senior Minister Lee Hsien Loong of Singapore warned the public about deepfake videos that falsely depicted him discussing international relations and promoting fraudulent investment schemes. The videos, hosted on TikTok, were used to misrepresent government positions and potentially deceive citizens. TikTok confirmed it removed the content, citing policies against misleading AI-generated media.
Two deepfake videos falsely depicted Senior Minister of Singapore Lee Hsien Loong commenting on US-China relations and the South China Sea. The videos misleadingly attributed views to him and the Singapore government. Posted on TikTok, the videos amassed over 190,000 views before being removed.
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 report also describes a prior deepfake incident where the minister's likeness was used to promote a fraudulent investment scam for financial gain.
The primary incident involves deepfake videos of a senior government official discussing sensitive international relations, which was intended to misrepresent state positions and influence public opinion at scale.
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 malicious actors intentionally deploying deepfake generation tools to create misleading videos of a public official.
Added categories peak at 3: Substantial (Epistemic Harm, Inferred)
Malicious actors deployed deepfake videos of Singapore Senior Minister Lee Hsien Loong on TikTok to misrepresent official government positions on sensitive geopolitical issues and promote fraudulent investment schemes. While the videos were quickly removed and caused minor immediate national security impact, they underscore the ongoing strategic challenge of synthetic media targeting state leadership and public trust.
The AI systems were used as tools by human creators to manipulate visual and audio content to generate deepfake videos.
Limited Risk: The report describes the creation of deepfakes, which are AI-generated content subject to specific transparency obligations under the EU AI Act.
The AI was used to manipulate visual and audio content to create realistic deepfake videos of a public figure speaking.
The report provides clear details about the deepfake videos, their content, view counts, and the platforms involved. The role of AI in generating these deepfakes is explicitly stated, and the response from the platform is documented.
None
Technology
Advanced Deepfake Generation
AI techniques allow realistic manipulation of visual and audio content.
Evolving Manipulation Quality
Rapidly evolving technology makes deepfakes increasingly harder to detect.
Data Inputs
Use of Authentic Video Footage
Real speeches and interviews were used as base data to train/alter videos.
Lack of Content Watermarking
No robust digital watermarks to verify authentic PMO videos.
Human Factors
Public Sharing of Deepfakes
Users sharing fake videos to warn others inadvertently increases their reach.
Difficulty in Visual Detection
Atypical facial movements and edits are hard for average viewers to spot.
Process and Methods
Delayed Platform Detection
Platform removed videos only after they gained significant views.
Inadequate Proactive Filtering
Lack of automated real-time prevention of high-profile political deepfakes.
Management
Reactive Content Moderation
Platform relied on post-upload reports and queries to take action.
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.
