Climate Action Group Posted Deepfake of Belgian Prime Minister Urging Climate Crisis Action

A deepfake video showing the Belgium’s prime minister speaking of an urgent need to tackle the climate crises was released by a climate action group.

Risk classification

• Primary risk domain: 4 Malicious actors
• Primary risk subdomain: 4.1 Disinformation, surveillance, and influence at scale

The incident involved the intentional creation and distribution of a political deepfake to influence public opinion and political discourse at scale.

Additional risk subdomains

• 3.1 False or misleading information: The deepfake video generated and spread false information by fabricating a speech from the Prime Minister.

Causal factors

• Entity: Human
• Intent: Intentional
• Timing: Post-deployment

The incident was caused by humans from Extinction Rebellion who intentionally created and distributed the deepfake video using post-deployment open-source AI tools.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The report describes the creation of a deepfake, which falls under Risk Level 3 as it requires transparency obligations to ensure users are aware they are interacting with AI-generated content.

AI system and alleged parties

• AI system: unspecified
• AI purpose: Deepfake Video Generation; Voice Generation
• Behaviour type: Tool
• Alleged developer: unknown
• Alleged deployer: Extinction Rebellion Belgium
• Alleged harmed parties: Shophie Wilmès, Belgian government

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Minor, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Minor, indirect Minor
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: The report explicitly describes the creation and spread of a deepfake video of the Prime Minister.

Directly caused: Extinction Rebellion created and distributed a deepfake video of PM Sophie Wilmès using machine learning to synthesize her voice and likeness.

Democracy

Reported: The report explicitly discusses how deepfakes raise serious questions about the integrity of democratic elections, policy-making, and public trust.

Directly caused: The deepfake of the Belgian PM was sent to cabinet and minister-presidents, potentially disrupting official communications and political discourse.

Indirectly caused: The incident contributes to the erosion of trust in public institutions and democratic processes.

Epistemic

Reported: The report explicitly describes epistemic harm through the spread of misinformation and the potential to mislead, misinform, and manipulate.

Directly caused: The deepfake fabricated a speech that the PM never gave, presenting false statements as real.

Indirectly caused: Such deepfakes contribute to a 'new infodemic' and erode the shared environment for facts.

People affected

• Occurrences reported: 1
• People reportedly harmed: 1
• People reportedly exposed: 4

Potential causes

Management

• Subjectivity of Online Platforms: Platform algorithms put reality at a disadvantage over engaging fakes.

Technology

• Advanced GAN Algorithms: GANs enable realistic creation of voice and likeness.
• Open Source Deepfake Tools: Software like DeepFaceLab allows anyone to create sophisticated deepfakes.
• Accessible Cloud Computing: Cloud computing provides the processing power needed to render deepfakes.

Data Inputs

• Availability of PM Video Footage: The group used a real address to the nation to train and modify the video.

Human Factors

• Intent to Pressure Policymakers: Activists used deepfakes to maintain climate pressure during lockdown.
• Susceptibility to Authority: Audiences easily believe fake narratives if presented by authority figures.

Process and Methods

• Ineffective Fact Checking: Efforts to correct fake content can accidentally strengthen its spread.
• Lack of Deepfake Detection Tools: Platforms lack reliable technology to automatically spot manipulated videos.

Regulatory Environment

• Lack of Global AI Regulation: No global consensus or cross-border ethical agenda exists for political AI.
• Absent Platform Standards: Online platforms lack consistent rules banning political deepfakes.

Information quality

• Classification confidence: High
• Reason for confidence: The reports clearly detail the creator, the target, the technology used, and the content of the deepfake. The context of the incident and its implications for democratic norms and misinformation are well-documented.

In April 2020, Extinction Rebellion Belgium released a deepfake video of Belgian Prime Minister Sophie Wilmes addressing COVID-19 and the climate crisis. While the incident did not cause physical harm or disrupt government functions, it demonstrates the evolving threat of AI-synthesized media to political communication, public trust, and democratic discourse.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: Belgium
• Alleged perpetrator: Extinction Rebellion Belgium

Threat characteristics

• Imminence: Long-term. The incident represents an ongoing strategic concern regarding the use of synthetic media in political discourse rather than an active, immediate crisis.
• Autonomy: Human-controlled. The AI tools were fully controlled by human operators who curated the training data, generated the video, and distributed it.
• Novelty: Evolved capability. While deepfakes existed, applying them to a sitting head of state during a national crisis (COVID-19) represents an evolved and highly targeted capability.

Impact by dimension

• Physical security: Negligible. No physical infrastructure, systems, or human safety elements were targeted or affected in this incident.
• Information security: Minor. Deepfake video of the Prime Minister created by a domestic activist group to advocate climate action; represents a minor information security incident manageable via standard public relations and security procedures.
• Sovereignty: Minor. Fabricated speech of the Prime Minister sent to government offices, representing a minor attempt to influence government discourse, but without causing disruption to core operations or constitutional processes.
• Economic security: Negligible. No economic systems, financial markets, or strategic technologies were targeted or compromised.
• Societal stability: Minor. Potential to erode public trust in institutions, but limited to a single campaign by an activist group with no widespread civil unrest or systematic rights violations.