Back to all blogs

What Are Shallowfakes? Everything You Need To Know About

What Are Shallowfakes? Everything You Need To Know About

Learn what shallowfakes are, how they differ from deepfakes, real-world examples, common risks, and practical ways to detect misleading images, videos, and audio.

By

By

Abhishek Kaushik

Abhishek Kaushik

Published on

Published on

Add Sherlock as a preferred source on Google

Shallowfakes can turn genuine recordings into misleading stories. A shortened speech may appear to contradict its speaker's intentions, while an old photograph can be recirculated as breaking news. Unlike deepfakes, which use artificial intelligence to generate or alter media, shallowfakes rely on simpler techniques such as cropping videos, changing playback speed, selectively editing audio, or adding misleading captions. These alterations can distort reality without advanced technology.

The risks associated with manipulated media are increasingly evident. According to a 2025 Gartner survey of 302 cybersecurity leaders, conducted across North America, EMEA, and Asia/Pacific from March to May 2025, 62% reported that their organizations experienced a deepfake-related attack involving social engineering or automated processes during the preceding 12 months. Although this statistic does not measure shallowfake prevalence, it highlights broader concerns about verifying digital content.

Understanding shallowfakes is essential because authentic media can become deceptive when edited or presented without context. Their accessibility and apparent credibility make them difficult to recognize, particularly when audiences encounter isolated clips or images without their original sources.

This guide explains what shallowfakes are, how they differ from deepfakes, common manipulation techniques, documented real-world examples, detection challenges, and practical prevention strategies. It also outlines how individuals and organizations can verify digital content before trusting or sharing it.

What Are Shallowfakes?

Shallowfakes are images, videos, or audio recordings that have been manipulated using relatively simple editing techniques or presented in a misleading context to influence how people interpret the content. They generally involve conventional editing or misleading contextual presentation rather than synthetic generation. The term is a useful descriptive label, not a universally standardized technical classification.

The original media used in a shallowfake may be completely authentic. The deception occurs when a small alteration changes the meaning of the content or encourages viewers to draw an inaccurate conclusion. For example, a video showing a public figure making a statement may be shortened to remove the sentence that explains its actual meaning. Similarly, an authentic photograph may be shared with a false date or location, making an unrelated event appear connected to a current issue.

Shallowfakes are particularly concerning because they do not necessarily require advanced technical skills or expensive software. Basic editing applications and social media tools can be enough to create misleading content that appears credible at first glance.

Their impact depends on how the content is altered, the context in which it is distributed, and the audience's willingness to verify it. Understanding what shallowfakes are is therefore the first step toward recognizing digital deception and distinguishing misleading media from content presented accurately and in context.

Image Showing What Are Shallowfakes

How Do Shallowfakes Work?

Shallowfakes work by changing how genuine digital content is presented, edited, or interpreted. Instead of generating an entirely new image, video, or voice, creators often modify existing material using basic editing software or remove information that would help viewers understand the original context. These changes can make an authentic recording appear to show something different from what actually happened.

The process typically involves selecting existing content, making a targeted alteration, and sharing it with a caption or explanation that encourages a particular interpretation. The following techniques illustrate how shallowfakes are commonly created.

1. Cropping and Trimming Videos

A video can be shortened to remove important statements, reactions, or events that explain what happened. Although the remaining footage may be genuine, viewers receive an incomplete version of the incident.

  • Removing the question that prompted a person's response can make an answer appear unrelated or inappropriate.

  • Cutting out the beginning or end of a recording can hide information that changes its meaning.

  • Sharing a short excerpt without linking to the complete video can prevent viewers from checking the original context.

2. Changing Playback Speed

Altering the speed of a video can make a person's movements or reactions appear unusual. Slowing down genuine footage may make an ordinary action seem suspicious, while speeding it up can make speech or behavior appear unnatural. Speed changes are routine in legitimate editing; they become misleading when an undisclosed adjustment materially changes the apparent behavior.

  • Slowing down a gesture can exaggerate its apparent duration or intensity.

  • Speeding up a recording can distort how quickly someone appears to react.

  • Sharing altered footage without disclosing the speed change can encourage misleading conclusions.

3. Editing or Rearranging Audio

Audio shallowfakes are created by removing words, combining separate statements, or rearranging portions of a genuine recording. These changes can make a speaker appear to express an opinion or intention they did not communicate in the original conversation.

  • Removing qualifying statements can make a cautious comment sound definitive.

  • Combining separate sections of a conversation can create a misleading sequence of statements.

  • Sharing a short audio clip without its surrounding conversation can obscure the speaker's intended meaning.

4. Cropping and Selectively Editing Images

Images can be misleading even when the underlying photograph is authentic. Ordinary adjustments for legibility or artistic effect are not necessarily deceptive. Cropping, selective editing, and changes to brightness or contrast can draw attention to certain details while hiding others.

  • Cropping a photograph can remove surrounding details that explain the scene.

  • Adjusting contrast or brightness can make a particular object or expression more noticeable.

  • Reusing an old photograph with a false date or location can make an unrelated event appear current.

5. Adding False Captions or Misrepresenting Context

Sometimes, the media itself is not edited at all. Instead, a misleading caption, headline, date, or description changes how audiences interpret it. This is one reason shallowfakes can be difficult to recognize through visual inspection alone.

  • An old video can be presented as footage of a recent event.

  • A genuine photograph can be falsely described as showing a different location.

  • A real statement can be paired with a caption that attributes a meaning the speaker did not intend.

Shallowfakes vs. Deepfakes: What Is the Difference?

Both shallowfakes and deepfakes can mislead audiences, but they usually operate differently. Shallowfakes typically reuse genuine media and alter its sequence, timing, framing, or description. Deepfakes commonly use machine-learning methods to synthesize or substantially transform a person's likeness, voice, or actions. Neither label by itself proves that a piece of media is deceptive: intent, disclosure, and context matter.

Feature

Shallowfakes

Deepfakes

Main method

Cropping, splicing, speed changes, selective editing, or false context

AI-based synthesis or substantial transformation of visual or audio content

AI involvement

Usually unnecessary; an AI-assisted editing tool does not automatically make an edit a deepfake

Generative or related machine-learning techniques are central to typical examples

Example

A real speech clipped before a crucial qualification

A synthetic video making a person appear to say words they never recorded

Verification focus

Find the longer original, compare edits, and establish event context

Examine provenance, compare against verified originals, and use forensic assessment where appropriate

Shared limitation

A technically genuine file can still be misrepresented

A convincing synthetic file may have no obvious visible artifacts

The distinction is not absolute. A misleading caption attached to an untouched photograph is often discussed alongside shallowfakes, although no pixels have changed. Conversely, routine AI-assisted noise reduction, subtitling, or color correction is not automatically a deepfake. The practical question is what was changed, whether the change was disclosed, and whether the resulting presentation distorts the underlying event.

Common Types of Shallowfakes

Shallowfakes can take several forms depending on the type of media and how it is presented. While the techniques described earlier explain how content is manipulated, the following categories highlight how misleading media appears and what viewers should verify.

Image Showing Types of Shallowfakes

1. Misleading Video Editing

Misleading video editing occurs when genuine footage is trimmed, rearranged, or altered to change how an event or statement is understood.

This can distort public discussions, damage reputations, and create confusion when viewers interpret an incomplete clip as an accurate representation of the original event.

For example, a video may be shortened to remove an explanation or qualification, making a speaker appear to support a position they were actually criticizing.

How to reduce the risk:

  • Locate and review the original, unedited video whenever possible.

  • Check whether important statements have been removed from the clip.

  • Verify the video's date, source, and surrounding context.

  • Compare the clip with reliable independent reporting before sharing it.

Reviewing the complete recording helps establish whether the edited version accurately represents what happened.

2. Misleading Image Manipulation

Image manipulation involves modifying or cropping genuine photographs to influence how viewers interpret a person, event, or situation.

Even when the underlying photograph is authentic, removing important visual details or presenting it with a misleading caption can create a false impression.

For example, an old photograph may be shared as evidence of a recent event, or a cropped image may omit details that change the meaning of the scene.

How to reduce the risk:

  • Use reverse image search to find earlier appearances of the photograph.

  • Verify the original publication date and location.

  • Compare the image with other available photographs of the same event.

  • Check whether the caption accurately describes what the image shows.

Image verification should consider both the photograph itself and the context in which it is being presented.

3. Misleading Audio Editing

Misleading audio editing occurs when parts of a genuine recording are removed, rearranged, or presented without the context needed to understand the original conversation.

This can make a person appear to have made a statement, promise, or admission that does not accurately reflect the full discussion.

For example, removing a question or qualification from a recorded conversation may change how listeners interpret the speaker's response.

How to reduce the risk:

  • Obtain the complete recording from a reliable source where possible.

  • Listen for abrupt cuts, missing sections, or unexplained changes in audio.

  • Compare the disputed excerpt with the original conversation.

  • Seek independent confirmation before relying on the recording for important decisions.

Audio inconsistencies alone do not establish manipulation, so findings should be assessed alongside the original material and other available evidence.

4. Misleading Captions and Context

Misleading captions occur when genuine digital content is accompanied by inaccurate descriptions, incorrect dates, false attributions, or explanations that distort its meaning.

This form of shallowfake can be particularly difficult to recognize because the original image or video may remain completely unchanged.

For example, footage from an earlier public event may be circulated with a caption claiming that it shows a newly occurring incident.

How to reduce the risk:

  • Verify the claim accompanying the image, video, or recording.

  • Check the date and location against credible sources.

  • Search for the original post or publication.

  • Avoid sharing content when its context cannot be established.

Accurate contextual information is essential because genuine media can still communicate false information when it is described incorrectly.

Real-World Examples of Shallowfakes and Illustrative Scenarios

Documented incidents demonstrate how modest changes to authentic footage can create major distortions. The first two cases below were examined by independent fact-checkers; the remaining examples are explicitly illustrative scenarios that show other potential techniques, not claims about verified incidents.

1. Documented Case: Nancy Pelosi Video Slowed to Suggest Impairment (2019)

In May 2019, a video from a Center for American Progress event circulated online with Nancy Pelosi's speech slowed down and the frame more tightly cropped. Posts and comments claimed she appeared intoxicated. FactCheck.org compared the viral recording with footage of the event and reported that the altered version added more than 40 seconds to the same segment. AFP's independent analysis likewise established that the footage had been slowed.

The underlying speaker and event were real. The deceptive impression arose from timing changes and the claim attached to the video, not from an AI-generated face or voice. Comparing the same passage at its original speed was essential to identifying the distortion.

2. Documented Case: Joe Biden Speech Cut Off Before Its Qualification (2020)

A clip from Joe Biden's March 7, 2020, campaign remarks circulated as though he had endorsed Donald Trump's re-election. The short version ended after the words about re-electing Trump, leaving out his qualification that Democratic infighting could produce that outcome. FactCheck.org and PolitiFact compared the excerpt with the longer remarks and explained the omitted context.

This case illustrates deception through omission rather than synthetic generation. The words in the short clip were spoken, but the edit removed the condition that explained their meaning. It also shows why an apparently authentic quotation should be checked against the full statement.

3. Illustrative Scenario: A Misleading Image Shared on Social Media

An authentic photograph of a small group at an event might be cropped to hide the larger gathering, or an older image might be captioned as though it shows a current incident. Neither scenario is presented here as a documented case. The mechanism is selective framing or false event attribution, which can make accurate pixels support an inaccurate narrative.

The distinction matters: an unaltered image paired with a false date requires contextual verification, while a cropped image also requires comparison with the wider original frame. Reverse image searches, contemporary reporting, and photographs from other angles can help reconstruct what the image actually depicts.

4. Illustrative Scenario: An Edited Screenshot Used to Spread False Information

Suppose a real business message is captured in a screenshot, then its date or a crucial line is edited to suggest that a payment was approved. That is manipulation of existing evidence. By contrast, a screenshot manufactured entirely from scratch, without any genuine underlying message, is fabricated evidence; it may be deceptive but is not necessarily a shallowfake under this article's working definition.

Screenshots are especially vulnerable because they strip away links, message history, and other context. A screenshot should therefore be treated as a lead for investigation rather than conclusive proof that the depicted communication occurred exactly as shown.

5. Illustrative Scenario: A Business Recording Used to Mislead Employees

A genuine meeting excerpt could be circulated to suggest that a manager approved a payment, authorized a policy change, or instructed employees to disclose sensitive information. The complete meeting might reveal that the proposal was conditional, hypothetical, or rejected. This is an illustrative workplace scenario, not a reported incident.

Organizations should confirm high-impact instructions through established channels and documented approval processes. A familiar voice, authentic meeting room, or credible-looking excerpt does not replace independent authorization.

6. Illustrative Scenario: An Audio Clip Rearranged to Change Meaning

A genuine conversation could be edited by joining nonadjacent statements so that a warning appears to be an endorsement. Unlike simply shortening a video, this changes the apparent order of the conversation. The recording may still sound natural because no synthetic voice is required.

Investigators would need to compare the excerpt with a longer recording, review the sequence of statements, and distinguish routine production edits from changes that materially misrepresent the speaker's intent. If the original is unavailable, the conclusion may remain uncertain.

Risks and Impacts of Shallowfakes

Shallowfakes matter because they can change how people interpret genuine information without requiring advanced technology or obvious signs of manipulation. A selectively edited video, misleading photograph, or audio recording taken out of context can influence public opinion, damage reputations, and create confusion. Since these techniques often rely on authentic material, viewers may trust the content without questioning how it has been presented.

As organizations and individuals increasingly depend on digital communication, the consequences of misleading media extend beyond social media misinformation. Shallowfakes can affect business decisions, political discussions, public trust, and the credibility of digital evidence.

1. Spreading Misinformation

Shallowfakes can distort information by removing important details or presenting genuine media with an inaccurate explanation. This can cause audiences to misunderstand events and form opinions based on incomplete evidence.

  • Short video clips can omit statements that explain a speaker's actual position.

  • Old photographs and recordings can be circulated as if they document recent events.

  • Misleading captions can encourage users to accept false claims about genuine images or videos.

  • Repeated sharing can allow inaccurate interpretations to spread before the original context is established.

2. Reputational Damage and Personal Harm

A misleading recording can affect how a person is perceived by colleagues, employers, clients, or the public. Even after the original context becomes available, the altered version may continue circulating.

  • An edited interview can make a candidate appear to give an inappropriate or misleading answer.

  • A selectively edited workplace conversation can create a false impression of an employee's conduct.

  • A cropped video of a public figure can make a statement appear more controversial than it was.

  • Incorrectly attributed images or recordings can undermine trust in an individual or organization.

3. Enabling Fraud and Deceptive Activities

Shallowfakes can also support fraudulent activities by making misleading claims appear more credible. Criminals may use genuine photographs, edited recordings, or selectively presented communications to create a false impression of authority or legitimacy.

  • An edited screenshot of a genuine payment confirmation can misrepresent its amount or approval status.

  • Genuine recordings can be presented without context to support fraudulent claims.

  • Misleading images may be used to create false impressions about products, services, or transactions.

  • A genuine message selectively cropped to omit a rejection can falsely imply managerial approval.

4. Influencing Political Debate and Democratic Participation

Manipulated media can affect public debate when genuine footage is presented in a misleading way. Political statements, protests, public events, and interviews can all be misrepresented through selective editing or false contextual information.

  • Shortened speeches can make political statements appear to mean something different from their original message.

  • Images from unrelated events can be used to support inaccurate claims about current developments.

  • Misleading edits can intensify disagreements or reinforce existing biases.

  • Rapid sharing can make it difficult for corrections to reach the same audience as the original content.

5. Undermining Trust in Digital Evidence and Content

Shallowfakes can create uncertainty about whether digital material accurately represents an event. This is particularly concerning when videos, photographs, or recordings are used in workplace investigations, journalism, legal proceedings, or organizational decision-making.

  • Edited recordings may omit information needed to understand an incident fairly.

  • Missing source information can make the authenticity and context of a file harder to establish.

  • Organizations may need to spend additional time checking the original material and its editing history.

  • Awareness of manipulation can lead people to dismiss genuine recordings without justification. File authenticity and accurate contextual presentation must be assessed separately.

6. Increasing the Burden on Verification Teams

When misleading content spreads, organizations may need to investigate its source, reconstruct its context, and determine whether an edit has changed its meaning. This can require additional time and resources, particularly when the original file is difficult to locate.

  • Investigators may need to compare several versions of the same recording.

  • Communication teams may need to respond to inaccurate claims and provide corrections.

  • Security teams may need to verify whether suspicious media was used to support fraud or impersonation.

  • Publishers may need stronger procedures for checking source material before distributing it.

Shallowfakes are therefore important not because every simple edit is deceptive, but because seemingly minor changes can significantly affect how digital information is understood. Recognizing their potential impact helps individuals and organizations evaluate media more carefully, verify its context, and avoid making decisions based on misleading presentations of genuine content.

Why Is Shallowfake Detection Difficult?

Identifying shallowfakes is challenging because the manipulated content may look completely authentic. Unlike many deepfakes, shallowfakes do not necessarily contain obvious visual glitches, unnatural facial movements, or synthetic audio patterns. Instead, they exploit editing choices and missing context, making verification dependent on examining the content's origin and how it has been presented.

Challenge

Why It Makes Detection Difficult

Authentic original material

Genuine images, videos, and audio can be misleading when selectively edited or presented without context.

Missing context

Cropping or removing sections can change the meaning without leaving visible signs of manipulation.

Misleading captions and descriptions

Authentic media can be paired with an incorrect date, location, or explanation.

Simple editing techniques

Basic editing tools can alter timing, framing, or sequencing without advanced technical skills.

Rapid online circulation

Misleading content can spread before its original context has been verified.

Limited access to original files

Investigators may have only compressed copies, screenshots, or reposted clips rather than original recordings.

Why Visual Inspection Alone Is Not Enough

A video may have natural lighting, realistic facial expressions, and clear audio while still misrepresenting an event. These qualities only indicate that the recording appears plausible; they do not establish that it has been presented accurately. Similarly, an image can pass basic visual checks while its caption or surrounding context remains false.

For this reason, shallowfake detection should combine content inspection with source verification, contextual research, and comparison against original material. Digital forensic tools may help identify edits or examine file properties, but their findings should be interpreted alongside other evidence.

The next section turns these limitations into a practical verification sequence. Finding an early upload can help trace circulation, but it does not necessarily establish the original recording or its true context.

How to Detect Shallowfakes?

Detecting shallowfakes involves verifying the original source, examining the complete recording, and checking whether edits or misleading context have changed its meaning. Because these manipulations often use genuine media, technical irregularities may be absent, making contextual verification particularly important.

Image Showing How to Detect Shallowfakes

1. Verify the Original Source

Start by tracing where the image, recording, or video appeared and seeking the original recording when available. The earliest discoverable upload is not necessarily the original source, and an official or credible account alone does not prove that the media or its context is accurate. Check whether the content comes from a credible publisher, an official account, or a source that can provide the original material. If the same clip appears across multiple platforms, compare the versions and look for differences in wording, duration, captions, or surrounding context.

2. Examine the Complete Content

Avoid drawing conclusions from short clips or isolated screenshots. Watch the full video, listen to the complete recording, or locate the original image wherever possible; sometimes the complete source is inaccessible, so conclusions must remain qualified. Pay attention to missing sections, abrupt transitions, changes in playback speed, and edits that remove important explanations. These signs do not automatically prove deception, but they can indicate that further investigation is needed.

3. Check Captions, Dates, and Context

A genuine photograph or video can become misleading when it is assigned the wrong date, location, or explanation. Distinguish the upload date from the date of the depicted event; confirm when and where the content was created, and compare the accompanying claims with reliable independent sources. Reverse image search can help locate earlier appearances of a photograph, while searches for distinctive phrases may reveal the original video or statement.

4. Look for Editing Inconsistencies

Review the material for unexplained cuts, inconsistent audio levels, abrupt changes in background noise, or visual transitions that interrupt the flow of an event. Where the original file is available, metadata and digital forensic tools may provide additional clues about its history, although metadata can be stripped, altered, or lost during sharing. Social media compression and ordinary editing can also create technical irregularities, so no single clue should be treated as conclusive evidence.

5. Cross-Check Important Claims

When a recording is being used to support a significant claim, look for corroboration from independent sources. For example, if a clip allegedly shows an executive approving a financial transaction, verify the instruction through an authorized communication channel rather than relying solely on the recording. In the same way, claims about public events should be checked against complete footage, credible reporting, or official records when available.

6. Preserve Evidence Before Sharing or Reporting

If you suspect that content has been manipulated, remember that a saved screenshot or downloaded copy can document what circulated without proving the original file is authentic. Retain the original link, capture relevant publication details, and preserve the file if you can do so lawfully. Avoid editing the evidence yourself, as doing so may make later examination more difficult. If the content involves fraud, impersonation, or a serious allegation, report it through the relevant platform or organizational process.

The key principle is to verify both the content and its context. A recording may be genuine while the message attached to it is misleading. Checking the original source, reviewing the complete material, and corroborating important claims can help identify shallowfakes before they cause further harm.

Conclusion

Shallowfakes demonstrate how easily genuine digital content can be manipulated to create misleading impressions. Through selective editing, cropped images, altered audio, and inaccurate captions, seemingly authentic material can distort facts, damage reputations, and weaken trust in digital information.

Addressing this challenge requires more than detecting visible signs of manipulation: the decisive issue is often the relationship between genuine content and the claim attached to it. Individuals and organizations must verify original sources, examine the full context, cross-check important claims, and preserve digital evidence when accuracy matters. Media literacy, responsible content sharing, and consistent verification procedures can help reduce the spread and impact of shallowfakes.

As digital content continues to shape personal, professional, and public decisions, understanding how shallowfakes work is an essential part of navigating information responsibly. The key is not simply to ask whether content is real, but whether it has been presented accurately and in its original context.

Abhishek Kaushik

Founder & CEO - Sherlock AI

Building fraud intelligence for live high-stake video interactions (financial discussions, hiring interviews, board meeting etc).

Catch candidate fraud before it costs you.

Sherlock AI helps you detect AI-assisted, proxy, and deepfake candidates, so every hiring decision is backed by evidence.

Ask AI about Sherlock.sh