Deepfakes and AI: Navigating truth in a digital age

artificial intelligence

By Aashi Tiwari

Video editing has been around for decades, allowing us t o tweak everything from blockbuster CGI to Instagram posts. But the rise of AI has dramatically changed the scale; it’s now easier than ever to edit — even create from scratch — incredibly convincing footage. Deepfakes, a form of what’s known as synthetic media, are created in this manner. What once required massive amounts of time, skill, and resources can now be produced in minutes by anyone with a computer and internet connection.

Deepfakes are AI-generated videos, images, or audio that look and sound real, but are entirely fake. They are created using algorithms trained on real media to mimic patterns in faces, voices, and movements, making the final output highly convincing. In practice, deepfakes can take many forms. From swapping faces to mimicking voices and expressions, their capabilities seem almost limitless.

The Dangers of Deepfakes

Like any other technology, deepfake algorithms can be harmless, valuable tools — or dangerous weapons. These videos have been used for election interference, war propaganda, morphed personal content, cyberbullying, and even blackmail. For example, a robocall impersonating former US President Joe Biden discouraged Democrats from voting in the New Hampshire primary, and a deepfake of President Zelensky urging Ukrainian troops to surrender in 2022 was circulated during wartime. These weren’t mere pranks — they were serious attempts to manipulate public opinion.

Due to their captivating and sensational nature, deepfaked videos and images can spread rapidly across the internet and social media. This makes the misinformation they propagate extremely difficult — and sometimes impossible — to contain. During the India–Pakistan tensions in May 2025, deepfakes spread so widely that even established news organisations mistakenly used them in their reporting. They can even present what experts call an "epistemic threat" — eroding public trust in legitimate news sources altogether.

How to Spot a Deepfake

If deepfakes are so dangerous yet increasingly common, how can we protect ourselves?

Your eyes and instincts are often your first — and best — line of defence. There are certain patterns, both visual and timing-related, that AI still struggles to replicate convincingly.

Visual signs to watch for:

*Unnatural facial expressions

*Abnormal blinking patterns

*Inconsistent lip movements

*Unusual texture or smoothness

*Mismatched lighting and shadows

Timing-related (temporal) clues:

# Jerky or stiff movement

# Poor transitions between frames

# Audio that’s out of sync with the video

For audio or videos with sound, additional red flags include:

- Vocal tone that differs from the speaker’s usual speech

- Lack or excess of background noise

- Incorrect timing of breathing or natural pauses

- Artificial background noise that feels forced or out of place

Even with AI's advanced capabilities, models still struggle to replicate the nuances of real media. Subtle elements — like how light reflects off skin or the fluidity of movement — can be especially hard to fake. When something seems “off”, it often is. Trust your instincts — those small imperfections are important clues.

The Role of Algorithmic Detection

While human observation is valuable, deepfakes are becoming increasingly difficult to spot with the naked eye. Algorithmic detection can be a powerful tool, employing advanced techniques to catch the subtle patterns that even trained humans may miss.

Some common algorithmic methods include:

* Anomaly detection: Identifies deviations from normal behaviour, such as irregular blinking or facial movement.

* Signature analysis: Detects patterns left behind by specific AI models — for example, unique pixel arrangements.

* ELA (Error Level Analysis): Examines compression levels across frames to identify regions that may have been altered — such as inconsistent compression on a face compared to the background.

Popular Deepfake Detection Tools

# Sensity AI: A user-friendly tool that takes a multilayered approach, analysing pixels, file structure, and voice patterns to deliver a comprehensive result.
https://sensity.ai/deepfake-detection/

# Deepfake-o-meter: An open-source platform that lets users upload content and run it through multiple detection models for comparative analysis.
https://zinc.cse.buffalo.edu/ubmdfl/deep-o-meter/landing_page

# Intel’s FakeCatcher: A real-time detection tool that analyses biological signals — particularly blood flow changes — across faces in a video.https://www.intel.com/content/www/us/en/research/trusted-media-deepfake-detection.html

A Particularly Alarming Misuse

Among the most disturbing uses of this technology is the creation of non-consensual deepfake pornography. This involves digitally superimposing someone’s face onto explicit content without their permission. It is a deeply invasive and harmful form of abuse, disproportionately targeting women and girls. The emotional and psychological impact can be devastating, leading to anxiety, trauma, and severe reputational damage.

Staying Safe in the Age of Deepfakes

While technology is developing to counter deepfakes, our strongest defence is not technological — it’s behavioural. Awareness is the first step: simply recognising that any video you watch could be manipulated is powerful. Next, apply visual scrutiny: trust your instincts and look for imperfections. Finally, when stakes are high, use detection tools and always cross-reference content with credible sources.

(Aashi Tiwari is a Grade 12 student at Inventure Academy, Bengaluru with a keen interest in cybersecurity and AI.)

Views are personal, and do not represent the stance of this publication.