This is a research prototype. The data and analyses are preliminary and not yet validated — we'd welcome your .
A Facebook page published an AI-manipulated image depicting Kosovo MPs Besnik Tahiri and Duda Balje in an inappropriate context. The image was identified as likely AI-generated by the detection tool WasitAi. Besnik Tahiri confirmed the image was fake and intended to damage his political reputation, providing the original photograph as evidence.
The Facebook page Specbuk reportedly published an AI-manipulated image allegedly suggesting inappropriate physical conduct between Kosovo MPs Besnik Tahiri and Duda Balje inside Kosovo's Parliament. KALLXO/Krypometër reported that the image had a high probability of being AI-generated, and Tahiri reportedly said the post was fake news intended to damage his political reputation.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
Using AI systems to gain a personal advantage over others such as through cheating, fraud, scams, blackmail or targeted manipulation of beliefs or behavior. Examples include AI-facilitated plagiarism for research or education, impersonating a trusted or fake individual for illegitimate financial benefit, or creating humiliating or sexual imagery.
Also classified as
The AI-generated image depicted the politicians in an inappropriate context, exposing social media users to misleading and potentially offensive synthetic content.
The deepfake generated and spread false visual information about the MPs, potentially misleading voters on the day elections were announced.
The incident involves the targeted manipulation of public perception and the defamation of specific politicians using a synthetic deepfake image to damage their reputation.
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 who intentionally generated and published a manipulated image to damage a politician's reputation after the AI model was deployed.
Added categories peak at 2: Minor (Epistemic Harm, Direct)
An AI-manipulated deepfake image of two Kosovo MPs was published on Facebook on the day early elections were scheduled, aiming to damage their political reputations. The image was quickly identified as fake using detection tools and original evidence, resulting in minor national security impact but highlighting the persistent threat of synthetic media in democratic processes.
The AI systems involved (the unnamed image generator and the WasitAi detector) operated as tools performing specific, well-defined functions (generating and detecting manipulated images) without autonomous planning or context retention.
Risk Level 3: Limited Risk. The incident involves AI-generated content (a deepfake image), which falls under the category of systems requiring transparency obligations to ensure users are aware they are interacting with AI-generated media.
The primary AI technology was used to generate a deepfake image of the politicians, while the secondary AI tool (WasitAi) was used to verify and detect the fabrication.
The report clearly details the specific deepfake post, the targeted individuals, the political context, and the fact-checking process using WasitAi. There is minimal ambiguity regarding the events, although the specific generator model is not named.
None.
Technology
AI Image Generation Tools
Generative AI was used to fabricate a realistic photo of MPs.
Data Inputs
Use of Real Photos as Input
Real public images of MPs were used to guide the AI generation.
Human Factors
Intent to Damage Reputation
Malicious actors used AI to politically harm political figures.
Process and Methods
Inadequate Platform Moderation
Facebook page Specbuk could post AI-manipulated content without pre-screening.
Regulatory Environment
Lack of Deepfake Regulations
No strong legal frameworks exist to penalize political AI manipulation.
AI system involved
Identified by automated extraction; can differ from the alleged developers above.
