Why In News?

Researchers at the University of California, Los Angeles (UCLA) developed an optical-neural processor that uses light to detect deepfake videos with nearly 98% accuracy.

What Are Deepfakes?

  • Definition: Deepfakes are audio, images or videos created or altered using artificial intelligence (AI) so that they appear real. They are often used to show people saying or doing things they never did.  

  • Technology: They are built with deep learning models such as Generative Adversarial Networks (GANs) and diffusion models, which learn to produce realistic faces and voices. 

  • Legal Term In India: Indian law calls such content “synthetically generated information” (SGI).  

Why They Are Hard To Spot

  • Human Limits: Recent research suggests that people correctly identify deepfakes only about half the time, no better than a coin toss.  

  • Rapid Improvement: New AI video tools produce ever more realistic content, which quickly outdates older detection methods. 

What Is The New Light-Based Detection Technique?

How It Works

  • Hybrid System: The system combines an electronic computer, which selects and encodes video frames, with an optical computer, where light rather than electronic circuits performs part of the computation.  

    • Free Mentorship Need Guidance for UPSC / WBCS / State PSC?
    100% Free & Confidential • Quick Callback

  • Spatial Light Modulator: The encoded features are displayed as a pattern on a programmable spatial light modulator. 

  • Light Diffraction: Laser light passes through this pattern and spreads as it travels. Sensors then measure the brightness at paired spots, and the difference in brightness indicates whether a video is real or fake.  

  • Parallel Screening: Unlike digital systems that check videos one by one, the system can analyse 15 or more videos at once in a single pass of light.  

Key Results

  • High Accuracy: In tests on 15 Celeb-DF videos at once, the system achieved an average accuracy of 97.79%, with 99.86% sensitivity (catching fakes) and 95.72% specificity (confirming real videos). 

  • Adapts To New AI Models: On unseen videos made with Google’s Veo 3, the system reached 94.80% accuracy after minimal fine-tuning.  

  • Harder To Fool: Under subtle tampering, the light-based versions were fooled about 9–14% of the time, while small all-digital detectors used for comparison were fooled every time.  

  • Energy Savings: The optical decoding step uses far less energy than an equally accurate digital decoder, though the digital front end still uses most of the power.  

Intended Role

  • First Line Of Defence: The researchers see it as a fast, high-volume first screening stage. Suspicious videos could then go to more detailed digital analysis. 

What Is Optical Computing?

Concept

  • Computing With Light: Optical computing uses photons (light) instead of electrons to process information.  

  • Speed And Efficiency: Light travels fast and can carry many signals at once without much heat, which makes optical computing attractive for AI tasks. 

  • Diffractive Networks: In diffractive optical networks, specially designed layers bend light to perform calculations as the light passes through them.  

Why It Matters For AI

  • Rising Energy Demand: AI models consume huge amounts of electricity in data centres, and optical processors could reduce this energy load.  

  • Real-Time Processing: Light-based processors can analyse large volumes of data quickly, which suits real-time tasks such as video screening. 

Deepfakes Challenges For India 

Threats To Society

  • Financial Fraud: Deepfake voices and videos are used to impersonate relatives, officials and celebrities to cheat people.  

  • Misinformation And Elections: Fake videos of leaders can spread false claims quickly and mislead voters. 

  • Harm To Women: Non-consensual deepfake imagery is used for harassment and blackmail, mostly targeting women.  

  • National Security: Deepfakes can be used to spread propaganda or fake orders during crises. 

Scale Of Cyber Fraud

  • Rising Losses: Indians lost ₹19,813 crore to fraud and cheating cases in 2025, with 21.77 lakh complaints on the National Cyber Crime Reporting Portal. 

What Is India’s Legal Framework On Deepfakes?

IT Rules Amendment, 2026

  • Legal Definition: The IT (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, define synthetically generated information for the first time in Indian law.  

  • Mandatory Labelling: AI-generated audio, images and videos must carry a prominent label and provenance metadata that cannot be removed. 

  • Faster Takedowns: Platforms must act on government or court orders within 3 hours (down from 36 hours), and remove non-consensual intimate imagery within 2 hours (down from 24 hours). 

  • Technical Safeguards: Platforms that enable AI content creation must deploy automated tools to detect and prevent unlawful synthetic content.  

Other Laws

  • Information Technology Act, 2000: Its provisions on identity theft (Section 66C), impersonation (Section 66D) and obscene content (Sections 67, 67A) apply to deepfakes. 

  • Bharatiya Nyaya Sanhita, 2023: The BNS covers cheating, defamation and the spread of false information.  

  • Digital Personal Data Protection Act, 2023: It protects personal data, including images and voice samples used to create deepfakes.

What Are The Key Challenges In Fighting Deepfakes?

Technological Arms Race

  • Detectors Fall Behind: Detection tools trained on older fakes often fail against content from newer AI models. 

    • Example: Even the UCLA system needed fine-tuning to handle videos made with Google’s Veo 3, which shows how each new generator can outpace existing detectors.  

  • Lab Versus Real World: Promising results in the lab may not hold up on compressed, low-quality videos shared on social media.  

    • Example: The UCLA results come from a bench setup with a laser, light modulator and camera, tested on benchmark datasets rather than live social media feeds.  

Enforcement Gaps

  • Speed Of Spread: A deepfake can go viral within minutes, long before any takedown takes effect.  

  • Cross-Border Creators: Many deepfakes are made and hosted abroad, beyond the reach of Indian law.  

  • Under-Reporting: Many victims, especially women, do not report deepfake abuse because of stigma.  

Rights Concerns

  • Free Speech Risks: Very short takedown windows may push platforms to remove lawful satire or parody to avoid liability. 

  • Low Public Awareness: Many users cannot recognise synthetic content or do not know how to report it. 

Way Forward

Technology-Led Solutions

  • Layered Detection: Fast optical screening can be paired with detailed digital analysis to check large volumes of content in real time.  

  • Content Provenance Standards: Digital “nutrition labels” can show where content came from and how it was edited.  

    • Example: The Coalition for Content Provenance and Authenticity (C2PA), formed in 2021 by companies including Adobe and Microsoft, has developed an open standard that attaches tamper-evident origin data to images and videos.  

  • Indigenous Research: India should fund home-grown detection tools. 

    • Example: Under the IndiaAI Mission, approved in 2024 with an outlay of ₹10,372 crore, the government has backed projects on deepfake detection and set up the IndiaAI Safety Institute to work on AI risks.

Legal And Institutional Measures

  • Protect Personality Rights: Courts and laws should protect a person’s face, voice and identity from misuse.  

    • Example: In Anil Kapoor vs Simply Life India (2023), the Delhi High Court protected the actor’s personality rights and restrained the use of AI to misuse his face and voice.  

  • Global Cooperation: India should work with other countries on common rules for cross-border deepfakes. 

    • Example: The EU AI Act requires that deepfakes be clearly disclosed as artificially generated, setting a transparency standard for AI-generated content.  

Awareness And Victim Support

  • Digital Literacy: Public campaigns should teach people to check sources and spot warning signs of fakes.  

  • Quick Reporting: Victims should use the 1930 cyber fraud helpline and the National Cyber Crime Reporting Portal immediately.  

Conclusion

UCLA’s light-based detector shows how optical computing could screen deepfakes faster and with less energy. Real protection, however, will need layered technology, strong laws like India’s 2026 IT Rules and an alert, digitally literate public.

Source: THEHINDU

PRACTICE QUESTION

Q. How should India balance free speech with the fight against AI-generated misinformation? (10 Marks, 150 Words)