The scene was set in a quiet home office in Hong Kong, where a finance worker logged onto what he believed was a high-stakes corporate video call. On his screen sat a grid of familiar faces, including the company’s Chief Financial Officer, whose gravelly, authoritative voice commanded the room. Alongside several directors, they discussed a confidential acquisition, eventually ordering the worker to authorize a series of massive wire transfers.By the time the call ended and his coffee had gone cold, $25.6 million was gone, not through a traditional server hack, but through a “hack of the human eye”. Every single “colleague” on that screen had been an AI-generated ghost, a digital phantom designed to steal.For over a hundred years, our society and justice systems have operated on a fundamental principle: seeing is believing. We relied on the objective truth of CCTV footage, recorded confessions, and leaked videos to anchor our reality.However, we have now entered the age of “Synthetic Truth”, a period where algorithms can reweave the very fabric of reality and the absolute authority of audio-visual evidence may be crumbling. In a world where any digital medium can be manufactured from nothing, how do we ever prove what actually happened?Behind the illusionAt the centre of this digital deception is a process known as a Generative Adversarial Network (GAN). To understand how these fakes become so convincing, one should imagine a high-stakes duel between an art forger, the “Generator,” and a museum curator, the “Discriminator”.The Generator uses vast amounts of public data, like photos and voice clips, to create a masterpiece, while the Discriminator looks for even the tiniest flaw, such as an unnatural shadow or a missing skin pore. This cycle repeats billions of times until the forgery becomes statistically indistinguishable from the real thing.In the early days of deepfakes, the “tells” were easy to spot; for instance, subjects famously never blinked because AI training datasets rarely included people with their eyes closed. However, developers quickly adapted, feeding thousands of blinking eyes into their databases to perfect the illusion overnight. Today, the combination of 5G networks and massive cloud computing means that criminals no longer need sophisticated laboratories. They can now alter live video streams on the fly, allowing for real-time impersonations during active security checks or high-stakes corporate meetings.The forensic fightbackAs these fakes approach perfection, human perception is no longer a reliable defence. We are forced to transition into an era of “Multi-Signal Forensics,” where verification systems act like a specialised jury looking for the “ghost in the machine”.One such method, Error Level Analysis (ELA), focuses on how images are saved. Every time a JPEG is re-saved, it loses data in a predictable way; if a face has been digitally spliced onto a body, the resulting compression patterns create a “digital scar” that is invisible to us but glaring to a computer.Furthermore, researchers are using sensor noise as a verification tool. Every physical camera sensor has unique manufacturing imperfections that leave a digital fingerprint, a “Photo Response Non-Uniformity” pattern, on every real photo. Synthetic images, by contrast, are “too clean” because they lack this physical noise.The social costThe danger of this technology extends far beyond financial fraud; it threatens the very foundation of social trust through what legal scholars call the “Liar’s Dividend”.As the public realises that everything can be faked, they may begin to assume that anything they find inconvenient or unpleasant is a fake. This creates a state of “reality apathy”, where citizens stop trusting any source of information and sink into a cynical void where truth is simply whatever aligns with their personal bias.In a courtroom, this could prove disastrous. A defence attorney might be presented with clear video evidence of a crime and simply dismiss it as an AI deepfake, making the standard of “beyond a reasonable doubt” an impossible bar to clear.The economic implications are equally terrifying; a convincing fake of a CEO making a forward looking statement could wipe out billions in market value before a verification team can even react.Architecture of truthWhile the Bharatiya Sakshya Adhiniyam, 2023 marks a significant leap by elevating electronic records to primary evidence and mandating SHA-256 cryptographic hashing under Section 63(4), the framework still contains structural gaps.The BSA’s authentication mechanism remains certificate-centric, validating how evidence was captured and stored, but provides no forensic standard for detecting whether that content is AI-generated or synthetically altered in the first place.Neither the BNS nor the BSA assigns criminal liability for the creation of deepfakes with intent to deceive a court, a lacuna that bad actors can exploit. What is urgently needed, therefore, is a dedicated statutory instrument, whether a standalone Digital Forensic Evidence Act or a targeted amendment to the BSA, that formally incorporates deepfake detection standards into the admissibility framework, establishes a notified National Digital Forensic Authority with the power to validate AI-generated evidence before it is tendered in court, and brings the certification requirement under Section 63 within reach of ordinary litigants, a concern the Supreme Court itself flagged but declined to resolve in Pune Bar Association v. Union of India (May 2026).We must recognise that deepfakes are an existential threat to the rule of law. If our legal systems cannot find a way to establish truth, they will drown in a sea of doubt.The goal must be to build an architecture of truth that is every bit as sophisticated as the machinery of the illusion.The writer, an IPS officer, is a Director General of Police in Tamil Nadu. Views are personalPublished on August 21, 2026