Alleged Use of Purportedly AI-Generated and Manipulated Media to Misrepresent Candidates and Disrupt Turkey's 2023 Presidential Election

During Turkey's 2023 presidential election, reportedly manipulated and allegedly AI-generated videos, audio, and images were used to smear candidates, purportedly link opposition figures to terrorist groups, and circulate a purported sex tape that reportedly contributed to presidential candidate Muharrem İnce’s withdrawal. These incidents reportedly misled voters, disrupted campaigning, and altered the electoral field.

Risk classification

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

The incident represents a coordinated, at-scale disinformation campaign using AI-generated deepfakes and manipulated media to influence the Turkish presidential election and manipulate public opinion.

Additional risk subdomains

• 4.3 Fraud, scams, and targeted manipulation: A targeted deepfake sex tape was used to humiliate candidate Muharrem Ince, forcing him to withdraw from the presidential race.

Causal factors

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

The incident was caused by human actors intentionally deploying AI deepfake and voice-cloning technologies post-deployment to manipulate public opinion and political outcomes.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Limited Risk: The incident involves AI-generated content such as deepfakes, which pose a moderate risk and require specific transparency obligations to ensure users are informed of their synthetic nature.

AI system and alleged parties

• AI system: unspecified AI
• AI purpose: Deepfake Video Generation; Voice Generation
• Behaviour type: Tool
• Alleged developer: Unknown voice cloning technology developers, Unknown generative AI developers, Unknown deepfake technology developers
• Alleged deployer: Supporters of Recep Tayyip Erdoğan, Government of Russia-aligned actors
• Alleged harmed parties: Truth, National security and intelligence stakeholders, Muharrem İnce, Kemal Kilicdaroglu, General public of Turkey, General public, Epistemic integrity, Electoral integrity, Democracy

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Substantial. 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 Substantial, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Minor, indirect Substantial
• Democracy: direct Minor, indirect Minor
• Privacy: direct Minor, indirect Negligible
• Psychological: direct Minor, indirect Negligible
• Epistemic: direct Minor, indirect Severe
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Malicious content

Reported: Yes, the reports describe the creation and viral spread of a fabricated sex tape and doctored political videos.

Directly caused: A fabricated sex tape targeting Muharrem Ince and a doctored video linking Kemal Kilicdaroglu to the PKK terrorist group were created and widely circulated on social media platforms.

Inferred additional harm: It is highly likely that numerous other malicious deepfakes, cheapfakes, and altered media were generated and shared across platforms like TikTok and Twitter during the 45-day election cycle.

Civil rights

Reported: Yes, the reports describe violations of political rights and the right to free and fair elections.

Directly caused: The political rights of Muharrem Ince were violated when a targeted smear campaign forced him to withdraw from the presidential race.

Indirectly caused: The civil rights of the Turkish electorate to receive accurate information and participate in a fair democratic process were compromised by the widespread disinformation.

Inferred additional harm: The systematic use of deepfakes to suppress opposition voices and manipulate voters likely led to broader violations of civil liberties and political participation rights across the country.

Democracy

Reported: Yes, the reports explicitly describe the erosion of democratic norms and voter misdirection.

Directly caused: The integrity of the Turkish presidential election was compromised, leading to the forced withdrawal of a candidate (Ince) and the manipulation of public opinion through fabricated media.

Indirectly caused: The dissemination of deepfakes deepened societal polarization and undermined public trust in democratic institutions, the media, and the electoral process.

Inferred additional harm: The successful deployment of these tactics in Turkey sets a dangerous precedent that likely encourages the future erosion of democratic norms in other global elections.

Privacy

Reported: Yes, the reports describe the non-consensual use of individuals' likenesses in fabricated media.

Directly caused: Muharrem Ince's privacy and likeness were violated through the creation of a non-consensual fabricated sex tape using his face.

Inferred additional harm: The potential use of personal data harvested from social media to target voters with deepfakes (similar to Cambridge Analytica tactics) suggests wider privacy risks, though not fully detailed.

Psychological

Reported: Yes, the reports describe severe distress and character assassination experienced by candidate Muharrem Ince.

Directly caused: Muharrem Ince suffered severe distress and public humiliation due to the release of a fabricated sex tape, describing it as a 'character assassination' and 'slander' that forced him to withdraw from the race.

Inferred additional harm: It is likely that other political figures and their families targeted by these deepfakes experienced significant psychological stress and anxiety, though specific details are not quantified.

Epistemic

Reported: Yes, the reports describe a torrent of AI-driven misinformation, fabricated videos, and the erosion of shared truth.

Directly caused: AI-generated deepfakes, such as the cloned voice of Kilicdaroglu speaking English and the fabricated sex tape, directly polluted the information ecosystem.

Indirectly caused: The prevalence of deepfakes created a 'liar's dividend' where real content could be easily dismissed as fake, making it extremely difficult for voters to discern truth from falsehood.

Inferred additional harm: The widespread erosion of shared reality likely caused long-term epistemic damage to the Turkish public's ability to trust any digital media or official communications.

People affected

• Occurrences reported: 1
• People reportedly harmed: 2
• People reportedly exposed: 1700000

Potential causes

Management

• Political Gain Over Truth: Candidates knowingly promoted debunked cheapfakes for campaign advantages.
• State-Sponsored Disinformation: Government directorates allegedly coordinated dark web deepfake campaigns.
• Media Control and Censorship: State control of 90 percent of media prevented objective deepfake reporting.

Technology

• Accessible Deepfake Generation Tools: AI tools allowed easy generation of realistic fake voices and videos.
• Algorithmic Search Bias: Google algorithms prioritized manipulated videos in search results.
• Dark Web Cyberwarfare: Deals with hackers to spread deepfakes and hack opposition sites.

Data Inputs

• Decontextualized Media Splicing: Splicing genuine footage with terrorist videos to create false narratives.
• Out-of-Context Training Data: Training AI on politician voices to generate unauthorized translations.
• Use of Foreign Pornographic Footage: Doctored footage from foreign websites used to fabricate character smears.

Human Factors

• Lack of Critical Thinking: Polarized voters accepted manipulated media without verifying its source.
• Exploitation of Cognitive Biases: Disinformation targeted voters' existing political and nationalistic biases.
• Public Susceptibility to Satire: Satirical AI content was taken out of context and believed to be real.

Process and Methods

• Overwhelmed Fact-Checking Teams: Small independent teams could not keep up with viral AI-generated content.
• Ineffective Content Verification: Platforms lacked real-time verification to stop deepfakes before going viral.
• Cryptographic Financing of Hacks: Using Bitcoin to pay dark web hackers for deepfake dissemination.

Regulatory Environment

• Weaponized Disinformation Laws: Laws were used to silence opposition journalists rather than curb deepfakes.
• Lack of Platform Accountability: Social media platforms complied with censorship requests, limiting truth.
• Absence of Foreign Interference Laws: Weak protections against state-backed Russian bot and troll campaigns.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide detailed, consistent, and cross-verified accounts from multiple reputable news agencies and fact-checking organizations (like Teyit) regarding the deployment of deepfakes and cheapfakes during the election, the candidates affected, and the political consequences.
• Ambiguities identified: Some reports conflate basic digital video editing (cheapfakes) with AI-generated deepfakes, such as the video shown at Erdogan's rally. Additionally, the exact technical tools used to generate the sex tape or clone Kilicdaroglu's voice are not specified.
• Alternative interpretations: The sex tape could be interpreted as a conventional smear campaign using traditional editing techniques rather than advanced AI, as its technical provenance was not independently verified by the reporting agencies.

During the 2023 Turkish presidential election, political actors deployed AI-generated deepfakes and voice clones to manipulate public opinion. The campaign successfully forced candidate Muharrem Ince to withdraw and falsely linked opposition leader Kemal Kilicdaroglu to terrorist groups, demonstrating a severe threat to democratic sovereignty and information security.

• Overall national security impact: Severe
• Scope: Single nation
• Primary target: Turkey
• Alleged perpetrator: Multiple political actors

Threat characteristics

• Imminence: Immediate. The incident unfolded rapidly during the critical final days of a presidential election campaign, requiring urgent response from fact-checkers and political actors.
• Autonomy: Human-controlled. The AI systems were utilized strictly as tools by human operators who generated, edited, and distributed the synthetic media.
• Novelty: Evolved capability. While deepfakes have been used previously, this represents a significant evolution in their political efficacy, directly forcing a presidential candidate to withdraw.

Impact by dimension

• Physical security: Negligible. No physical systems, critical infrastructure, or kinetic threats were targeted or affected in this incident.
• Information security: Severe. AI-generated deepfakes and voice clones were deployed at scale to target presidential candidates, polluting the information ecosystem and severely manipulating public opinion during a national election.
• Sovereignty: Severe. The integrity of a core sovereign democratic process was compromised, directly leading to the forced withdrawal of candidate Muharrem Ince and the manipulation of voters.
• Economic security: Negligible. No significant economic infrastructure, financial systems, or strategic technologies were targeted or compromised.
• Societal stability: Substantial. The widespread dissemination of synthetic media deepened societal polarization, eroded public trust in democratic institutions, and violated the privacy rights of targeted individuals.