Deepfakes — When You Cannot Believe What You See

In 2024, a finance worker at a multinational company in Hong Kong received a video call from his chief financial officer. The CFO, appearing live on screen alongside several other senior colleagues, instructed the worker to transfer 25 million US dollars to a series of bank accounts. The worker complied. Every person on that video call — every face, every voice, every gesture — was a deepfake. None of them were real. The money was gone.

A deepfake is a piece of media — usually a video or audio recording — that has been created or manipulated using artificial intelligence to make it appear that someone is saying or doing something they never actually said or did. The technology uses a type of AI called deep learning to analyse thousands of images and audio samples of a real person, learn the patterns of their face, voice, and movements, and then generate convincing synthetic versions.

The technology has improved with extraordinary speed. Early deepfakes, which first appeared around 2017, were often easy to spot — the faces looked slightly wrong, the mouth movements did not quite match the audio, the eyes behaved strangely. Today, the best deepfakes are virtually indistinguishable from genuine footage, even to trained experts using forensic analysis tools. What once required expensive equipment and specialist skills can now be produced using free software on an ordinary laptop.

This combination of increasing quality and decreasing cost has made deepfakes one of the most significant technological threats of the decade — not because the technology itself is inherently evil, but because it attacks something fundamental to how human societies function: the ability to trust what we see and hear.

The most immediate and widespread harm caused by deepfakes has been in two areas: non-consensual intimate imagery and political disinformation.

The overwhelming majority of deepfakes currently circulating online — estimated at over 90% — are non-consensual intimate content, in which the faces of real people, almost always women, are placed onto explicit material without their knowledge or consent. The victims include both public figures and ordinary individuals, and the psychological harm can be devastating.

Political deepfakes represent a different but equally serious threat. During election campaigns in multiple countries, deepfake videos have been used to make candidates appear to say things they never said, to fabricate endorsements, and to create false impressions of events that never occurred. In 2023, a deepfake audio clip of a political leader appeared to show him making inflammatory statements — it was shared millions of times before it was identified as fake.

The danger of political deepfakes is not only that people will believe false content. It is also what researchers call the "liar's dividend" — the idea that in a world where any video or audio could be fake, real evidence of genuine misconduct can be dismissed as a deepfake. A politician caught on camera doing something wrong can now claim the footage is AI-generated. The existence of deepfake technology does not just make lies more convincing — it makes truth less powerful.

The fight against deepfakes mirrors the fight against doping in sport — it is an arms race between those who create fakes and those who try to detect them.

Early deepfake detection relied on visible flaws — unnatural blinking patterns, inconsistent lighting, distorted edges around the face. As deepfake technology improved, these flaws disappeared. Modern detection tools use AI to analyse subtle features that the human eye cannot perceive — micro-expressions, blood flow patterns beneath the skin, the way light reflects off the surface of an eye.

Some organisations are taking a different approach: rather than trying to detect fakes after they are created, they are working to verify authentic content at the point of creation. The Content Authenticity Initiative, backed by Adobe and Microsoft, is developing a system that embeds a digital signature into photos and videos at the moment they are captured. Any subsequent manipulation would break the signature.

Governments are also responding with legislation. The European Union's AI Act requires that AI-generated content be clearly labelled. China has introduced laws requiring deepfake creators to obtain consent and to watermark synthetic content. Several US states have passed laws specifically criminalising malicious deepfakes.

But enforcement remains the central challenge. Deepfakes can be created anywhere and distributed instantly across borders. A video created in one country can cause harm in another before any legal system has time to respond. Technology, once again, is moving faster than the law.

The deepest challenge posed by deepfakes is not technical — it is philosophical. If any piece of media can be faked, how do we decide what is real?

Some researchers argue that we are entering a "post-truth" era in which objective, verifiable reality is being eroded — not just by deepfakes, but by a broader information environment in which social media algorithms prioritise engagement over accuracy, and people increasingly consume information that confirms their existing beliefs.

Others are more optimistic. They point out that every major communication technology — the printing press, photography, radio, television — was initially accompanied by fears that it would be used to deceive and manipulate. Each time, societies eventually developed the tools, institutions, and cultural norms needed to distinguish reliable information from unreliable. Deepfakes, they argue, will follow the same pattern.

What most experts agree on is that media literacy — the ability to critically evaluate the information we consume — has become an essential life skill. Understanding how deepfakes work, knowing where to check the authenticity of content, and developing the habit of pausing before sharing something online are no longer optional skills. They are as fundamental as reading and writing.

The technology behind deepfakes is not going away. It will continue to improve. The question is not whether we can stop it — we cannot — but whether we can build a society that is resilient enough to function in a world where seeing is no longer believing.