Deepfakes and Synthetic Media

Synthetic media is digital content that is generated or substantially altered by algorithms. A deepfake is synthetic audio, video or imagery created using artificial intelligence, especially deep learning, to make a person appear to say or do something that did not occur, or to imitate the person's face or voice convincingly. The term describes the technique and does not necessarily imply malicious intent.

How deepfakes are created

AI models learn facial, vocal or behavioural patterns from training data and then generate, reconstruct or replace selected features in digital content. Techniques may include autoencoders, generative adversarial networks and diffusion models.

• A face swap replaces one person's facial features while retaining the movements of another person.
• Lip synchronisation alters mouth movements to match fabricated or translated speech.
• Voice cloning generates speech resembling a target person's voice from recorded samples.
• Deepfakes differ from simple edited media because AI learns and reproduces complex patterns rather than merely cutting or rearranging content.

Applications and risks

Synthetic media can support entertainment, dubbing, education, accessibility and creative production. Its misuse creates serious identity and information integrity risks.

• Impersonation can facilitate fraud, reputational harm and harassment.
• Non-consensual intimate imagery violates privacy and dignity.
• Fabricated political or social content can spread misinformation and weaken public trust.
• Increasing realism makes unaided visual detection unreliable and can also cause genuine recordings to be falsely dismissed.

Governance and safeguards

Effective governance combines consent, platform accountability, technical provenance, detection tools and media literacy. In India, conduct involving deepfakes may attract existing law depending on its purpose and effects.

• Section 66D of the Information Technology Act, 2000 applies to cheating by personation using a communication device or computer resource.
• The Information Technology Rules, 2021 impose due diligence and grievance redressal obligations on intermediaries.
• Watermarking, content credentials and traceable provenance can disclose AI generation, while forensic tools can help assess authenticity.
• Governance must distinguish legitimate creative use from deception, fraud and non-consensual manipulation.

How UPSC asks this