This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A deepfake video of Vice President Kamala Harris, which used AI-manipulated audio to make her appear to disparage President Joe Biden and herself, was amplified by Elon Musk on X. Although the original creator labeled the content as a parody, Musk reposted it without any disclaimer, leading to widespread circulation and debate regarding X's policies on synthetic media and election integrity.
The X user @MrReaganUSA uploaded a deepfake of Kamala Harris saying damaging comments about Joe Biden and her own qualifications for the presidency, originally marking it as a parody. The post was shared and amplified eight hours later via Elon Musk's account without the disclaimer.
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 video refers to Harris as a 'diversity hire' and mocks her race and gender, representing unfair misrepresentation.
The video generates false and misleading audio of a public official making statements she never actually made.
The incident involves the use of AI-generated synthetic voice to mimic a political candidate, which was shared at scale to influence public opinion and the democratic process during an election campaign.
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 primarily caused by human actions, specifically the intentional creation of the altered video by a user and its subsequent intentional amplification by Elon Musk on X without appropriate labeling.
Added categories peak at 4: Severe (Epistemic Harm, Indirect)
An AI-generated deepfake video of Vice President Kamala Harris went viral after being amplified by Elon Musk on X. The incident highlights the substantial threat of synthetic media to democratic elections and information security, emphasizing the challenges of platform policy enforcement and regulatory gaps regarding AI-generated political content.
The AI system functioned as a tool to generate synthetic voice and manipulate video based on direct human inputs. It did not exhibit autonomous decision-making or goal-directed behavior, but rather executed a specific, well-defined function of voice replication.
Risk Level 3: Limited Risk. The report describes an AI-generated deepfake video, which falls under the category of AI systems that pose a moderate risk and require specific transparency obligations, such as notifying users that the content is AI-generated.
The AI system was used to generate a synthetic voice mimicking Kamala Harris's voice, which was then paired with edited video clips to create a deepfake video.
The reports provide consistent, detailed accounts of the incident, including the specific content of the deepfake, the platform where it was shared, the view counts, and the reactions from various stakeholders. The role of AI in generating the voice is clearly established, and the primary debate centers on platform policy enforcement rather than factual disputes about the event.
None
Technology
Generative AI Voice Cloning
AI tools allowed highly realistic manipulation of Harris's voice.
Deepfake Creation Tools
Accessible technology enabled the creation of deceptive political video.
Data Inputs
Original Campaign Video Audio
Unaltered campaign audio was manipulated to generate the fake voiceover.
Lack of Digital Watermarks
Absence of metadata or watermarks allowed the video to circulate unlabeled.
Human Factors
User Sharing Without Verification
Musk shared the deepfake to his 191M followers without verifying or labeling.
Audience Susceptibility
Voters struggle to distinguish AI-generated content from reality.
Process and Methods
Flawed Community Notes Process
X's crowdsourced moderation failed to append a warning note in a timely manner.
Inconsistent Policy Enforcement
X failed to enforce its misleading media policy on a high-profile user.
Management
Prioritizing Engagement Over Safety
Platform owner prioritized viral engagement over enforcing safety guidelines.
Lack of Executive Accountability
Leadership bypassed platform rules without facing internal enforcement actions.
Regulatory Environment
Ambiguous Election Laws
The 1971 Federal Election Campaign Act is ambiguous regarding AI technology.
Gridlocked Regulatory Bodies
FEC split along party lines, failing to pass rules on deceptive AI ads.
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.
