File0001

Deepfakes Future Threat Landscape Deepfakes Guidelines: Mitigating risks while fostering innovation This section outlines the importance of AI ethics in deepfake technology development. Overall, developers should ensure deepfake technology adheres to ethical standards, safe­ guards personal data, and prevents misuse. Deepfake technology must comply with laws, re­ spect social ethics, and align with correct values. Technology should enable content authenticity verification and prevent the misuse of synthetic media, promoting responsible and trustworthy use. This is paramount due to the profound societal impact of deepfakes and help steer the devel­ opment of deepfakes towards positive and constructive uses while mitigating associated risks. Developers should thus adhere to all the points in the below list when developing and managing deepfake technologies. 5. Guidance for Deepfake Technology Developers 5.1.1 Adhere to all relevant local and international data privacy laws, such as GDPR, CCPA, and KSA’s PDPL and Anti-Cyber Crime Law, and integrate legal compliance checks into your development environment (e.g., policy-as-code frameworks such as Terraform and Cedar Policy Language) (PDPL Article 3, AI Ethics Principles – Compliance Section, page 32) (PDPL Article 3, AI Ethics Principles – Compliance Section, page 32) 5.1 Regulatory Compliance Deepfakes Guidelines: Mitigating risks while fostering innovation 5.2.1 Implement robust data protection measures, including secure data transfers, and limit data collection to the minimum necessary. Apply privacy-preserving techniques like anonymization and conduct a Data Privacy Impact Assessment if data is used beyond contractual relationships. 5.2.2 Integrate privacy by design in technology development, ensuring minimal data retention and the use of privacy-preserving techniques like anonymization to reduce re-identification risks. Establish automated review cycles to securely delete unnecessary data (e.g., using cryp­ tographic erasure), minimizing risks of unauthorized access 5.2.3 Implement consent management systems within AI tools, ensuring data is used only with proper consent. Record and manage consent transactions, with clear documentation of con­ tractual relationships when applicable 5.2.4 Allow individuals to request the removal of their likeness or personal data from deepfake training datasets, with clear processes to prevent nonconsensual use. 5.3.1 Document every aspect of your AI model, including data sources, preprocessing steps, algorithmic architecture, and the decision-making process. Ensure this documentation is ma­ chine-readable and integrated into your CI/CD pipeline for continuous updates. 5.3.2 Incorporate explainability features within AI models using techniques such as LIME (Local Interpretable Model-agnostic Explanations ) or SHAP (SHapley Additive exPlanations ) to ensure that users and stakeholders can understand how outputs are generated (PDPL Article 19, AI Ethics Principles – Principle 2) (PDPL Articles 4,19 and 28; AI Ethics Principles – Principle 2) (AI Ethics Principles – Principles 5 and 7) 1 A series of automated steps that helps software teams deliver code faster, safer, and more reliably. 2 A technique that approximates any black box machine learning model with a local, interpretable model to explain each individual prediction. 3 SHAP (SHapley Additive exPlanations) is a game theoretic approach to explain the output of any machine learning model. (PDPL Articles 4 and 28; AI Ethics Principles – Principles 2) (AI Ethics Principles – Principle 6) 5.2 Data Privacy and Protection 5.3 Transparency and Explainability Deepfakes Guidelines: Mitigating risks while fostering innovation 5.3.3 Develop and implement non-intrusive digital watermarking techniques (e.g., stegano­ graphic methods) and ensure these markers are resistant to common attacks (e.g., re-compres­ sion, cropping, and filtering)