Back to all blogs
Learn how deepfakes and shallowfakes differ, how they are created, their risks, and ways to identify manipulated media and reduce its misuse.
Digital manipulation makes it increasingly difficult to determine whether an image, video, audio recording, or document accurately represents reality. Two prominent forms are deepfakes and shallowfakes. Both can mislead audiences, but they differ in their creation methods, technical complexity, and the evidence needed to investigate them.
Deepfakes use artificial intelligence (AI), often generative models, to synthesize or substantially alter realistic media. They can depict public figures making invented statements, reproduce an executive's voice, or create a synthetic face in a video call. Shallowfakes instead commonly alter genuine material through cropping, selective splicing, playback-speed changes, or misleading captions. They may need no generative AI at all, yet they can radically change the apparent meaning of an event.
The distinction matters amid widespread concerns about unreliable information. The World Economic Forum's Global Risks Report 2026 ranked misinformation and disinformation second in its two-year global-risk outlook, based on its survey of more than 1,300 experts. That ranking addresses information risks broadly, rather than measuring the prevalence of either deepfakes or shallowfakes specifically.
Deepfake-related impersonation is also a documented organizational concern. In Gartner's September 2026 report, 41% of 297 surveyed senior cybersecurity leaders reported at least one social-engineering incident involving a deepfake during an employee audio call in the previous 12 months; 36% reported such an incident during a video call. These are survey reports of incidents, not figures for financial losses, and they do not measure shallowfake prevalence.
A convincing face or voice is not automatic proof of authenticity; neither is a real video necessarily a fair account of what happened. Understanding both categories enables individuals and organizations to identify suspicious content, distinguish synthetic fabrication from misleading editing, and select appropriate verification safeguards. The sections below compare how each technique works, what they share, where their risks diverge, and how to respond without treating every edit as deception.
What Is a Shallowfake?
A shallowfake refers to authentic media that has been manipulated through relatively simple editing techniques or presented in a misleading context to create a false impression. It typically relies on authentic underlying material rather than generating an invented face, voice, or event. Cropping, rearrangement, selective omissions, and false contextual claims can make truthful source material appear to support an untrue conclusion. The label is a useful descriptive distinction, not a universally standardized technical category.

How Are Shallowfakes Created?
Shallowfakes can be produced using readily available editing software and basic media manipulation techniques. A person might cut out a sentence from a longer speech, combine clips from different events, change the playback speed of a video, or attach a misleading caption to an authentic photograph. These changes may not alter the original person's face or voice, but they can substantially change how the content is interpreted.
For example, a genuine video of a company executive discussing a financial challenge could be edited to remove the explanation that follows, making the executive appear to announce a business failure. Similarly, an authentic recording played at a different speed could make someone's speech sound unusual or misleading. In both cases, the original media exists, but its presentation distorts the intended meaning.
Common Examples of Shallowfakes
Misleading video edits: Removing parts of a speech to change its meaning or make a statement appear out of context.
Altered playback speed: Slowing down or speeding up footage to make a person appear intoxicated, confused, or unusually aggressive.
Misleading captions: Attaching text that inaccurately describes what a real image or video shows, such as falsely identifying a speaker or action.
Selective audio editing: Cutting out words or combining audio fragments to misrepresent what someone said.
Context manipulation: Reusing authentic material with an incorrect date, place, or event attribution, even if no pixels or audio are changed.
Why Can Shallowfakes Be Difficult to Identify?
Shallowfakes can be convincing because the underlying photograph, video, or recording may be authentic. Viewers may focus on the apparent evidence without checking whether it has been edited or presented in the correct context. Unlike some AI-generated media, shallowfakes may contain no obvious visual artifacts or synthetic features.
In short, shallowfakes demonstrate that digital deception does not always require advanced AI. Simple edits and misleading context can be enough to distort genuine information and influence public perception.
What Are Deepfakes?
Deepfakes are digitally manipulated or synthetically generated images, videos, and audio recordings created using artificial intelligence (AI). They commonly involve deep learning techniques that enable systems to imitate a person's face, voice, expressions, or movements. Unlike shallowfakes, which typically rely on basic editing or misleading context, deepfakes can generate realistic content that makes a person appear to say or do something that never happened.
Deepfake technology has legitimate applications in filmmaking, visual effects, entertainment, accessibility, and media production. However, when used without consent or with deceptive intent, it can enable impersonation, misinformation, identity fraud, and reputational damage.

How Are Deepfakes Created?
Deepfakes are produced using machine-learning systems that synthesize or modify images, video, or audio. Some approaches train or fine-tune models on substantial recordings of a target, but that is not universal: newer systems may work from shorter voice samples, a few reference images, or pretrained generative models. Depending on the technique, the system may create a face, reenact expressions, or generate speech resembling an identifiable person.
For example, a face-swapping model can replace the face of a person in a video with another person's likeness. Voice-cloning systems can generate speech that sounds similar to a real person's voice, while other generative models can create entirely synthetic images or video sequences. The resulting media may appear convincing, particularly when viewers see only a short clip or have limited access to the original source.
Common Examples of Deepfakes
Face swapping: Replacing a person's face in an existing video with someone else's likeness.
Voice cloning: Generating speech that imitates a person's voice to make false statements or impersonate them.
AI-generated video: Creating realistic footage of a person appearing to speak, move, or perform an action that never occurred.
Facial reenactment: Manipulating facial movements or expressions to make a person appear to communicate different words or emotions.
Deepfake-generated identity elements: Using AI-created profile faces or cloned voices within an impersonation attempt. Synthetic identity fraud can involve such elements, but a synthetic identity is not automatically a deepfake.
Why Are Deepfakes a Growing Concern?
Deepfakes can be especially dangerous when they imitate trusted individuals, such as business executives, public officials, family members, or job candidates. A fabricated video call or cloned voice message may be used to pressure someone into transferring money, disclosing confidential information, or accepting a false claim as genuine.
This challenge is distinctive because convincing synthetic media can present seemingly direct evidence of a person speaking or appearing somewhere without that event having occurred. High-quality output may contain no visible glitches, while genuine low-quality footage may look suspicious. Authentication systems that rely on recognizing a voice or face alone can therefore be vulnerable. The specific consequences, including fraud and reputational harms, are examined below.
Deepfakes differ from shallowfakes primarily in the methods used to create or manipulate media. Understanding these methods helps individuals and organizations recognize the different risks and choose appropriate verification measures.
Deepfakes vs. Shallowfakes: Key Differences
Deepfakes and shallowfakes may both distort reality, but they differ in how digital content is created, manipulated, and used to mislead audiences. Understanding these differences helps individuals and organizations recognize potential deception, assess the credibility of digital media, and choose appropriate verification methods. The table below compares the two techniques across key factors, including production methods, the role of AI, realism, accessibility, detection challenges, and potential misuse.
Comparison Factor | Deepfakes | Shallowfakes |
|---|---|---|
Definition | Digital media generated or manipulated using AI to imitate real people, voices, or events. | Existing media altered through basic editing or presented in a misleading context. |
Production Methods | Often use deep learning, generative AI, voice cloning, or face-swapping technology. | Commonly use cropping, splicing, playback-speed changes, selective editing, or misleading captions. |
Role of AI | AI is fundamental to generating or substantially manipulating the media. | Generative AI is not required, although AI tools may assist with conventional editing. |
Technical Complexity | Can require specialized tools and technical expertise, depending on the method. | Usually easier to create using widely available editing software. |
Source Material | May generate entirely synthetic content or modify existing images, videos, and audio. | Usually relies on existing media that is edited or taken out of context. |
Realism | Can convincingly imitate a person's appearance, voice, expressions, or movements. | Often appears authentic because the underlying footage or image may be genuine. |
Common Examples | Cloned executive voices, face-swapped videos, synthetic video calls, and AI-generated speeches. | Edited speeches, misleading video clips, altered playback speed, and false captions on authentic images. |
Verification Challenges | Synthetic content can look realistic; technical assessment must be paired with provenance and corroboration. | Genuine content can be misleading; check the complete source and the accuracy of its context. |
Accessibility and Cost | Tools are increasingly accessible, although advanced results may require more resources. | Generally inexpensive and straightforward to produce. |
Potential Misuse | Identity impersonation, financial fraud, fabricated endorsements, and misinformation. | Misleading political claims, reputational damage, deceptive narratives, and misinformation. |
Overlap Between Techniques | AI-generated footage can also be cropped or misleadingly captioned. | AI-assisted cropping or cleanup is not automatically a deepfake; classification depends on what changed. |
Key takeaway: Deepfakes primarily involve AI-generated or AI-altered media, whereas shallowfakes typically rely on simpler edits or misleading context. Both can deceive audiences, so verifying the original source and context is essential.
What Are the Similarities Between Deepfakes and Shallowfakes?
Although deepfakes and shallowfakes differ in how they are created, both can influence how audiences interpret digital content. Deepfakes generate or alter media using artificial intelligence, while shallowfakes typically rely on conventional editing techniques or misleading context. Despite these technical differences, both can make inaccurate claims appear credible by exploiting the way people interpret visual and audio evidence.

Both Can Make Misleading Claims Appear Credible
Deepfakes and shallowfakes can create a gap between what digital content appears to show and what actually happened. A deepfake may fabricate a realistic video of a person making a statement they never made, while a shallowfake may use genuine footage but remove important context that changes the statement's meaning.
In both cases, audiences may interpret the presented material as reliable evidence without recognizing how it has been manipulated. The similarity lies in their ability to give misleading claims an appearance of authenticity, even though the underlying methods are different.
Both Can Misrepresent Identity and Communication
Both forms of manipulation can create misleading impressions about a person's identity, statements, or intentions. Deepfakes may reproduce someone's facial appearance or voice, while shallowfakes may selectively edit genuine conversations, photographs, or recordings to misrepresent what a person communicated.
For example, an AI-generated voice recording may falsely suggest that an executive issued an instruction, while a selectively edited recording may distort the meaning of an actual executive's statement. Although one involves synthetic content and the other alters authentic material, both can make a communication appear more authoritative or conclusive than the available evidence supports.
Both Depend on Context and Audience Interpretation
The effectiveness of deepfakes and shallowfakes depends partly on how audiences encounter and interpret the content. A realistic synthetic video may appear trustworthy because it resembles an authentic recording, while a shallowfake may seem credible precisely because its underlying footage is genuine.
Captions, distribution channels, surrounding narratives, and audience expectations can further influence how the material is understood. When content is shared without sufficient background information, viewers may accept its apparent meaning without recognizing the manipulation.
These similarities show why neither realistic appearance nor authentic source material is sufficient to establish the accuracy of digital content. Both deepfakes and shallowfakes require careful verification of the source, context, and underlying claims.
Risks of Deepfakes and Shallowfakes
Both deepfakes and shallowfakes can create serious risks for individuals, businesses, and society. While deepfakes can enable convincing AI-powered impersonation, shallowfakes can distort authentic content through simple edits or misleading context. The impact depends on how the manipulated material is used, how widely it spreads, and whether people rely on it when making decisions.
The Risks of Shallowfakes
Shallowfakes are relatively easy to create and distribute because they often require only basic editing tools. A misleading caption, selectively edited video, or altered recording can influence how an event is perceived without generating any synthetic media.
Common risks include:
Misinformation and disinformation: Genuine images or recordings may be presented with false claims or misleading context.
Reputational damage: Edited clips can make individuals or organizations appear to have said or done something differently from what actually occurred.
Political manipulation: Selectively edited speeches or videos can distort public understanding of political events.
Fraud and deception: Altered documents, screenshots, or recordings may be used to support false claims or fraudulent requests.
Rapid circulation: Short, emotionally charged clips can spread quickly before viewers check their authenticity or context.
The main challenge is that shallowfakes may contain authentic material, making it necessary to investigate how that material has been edited or presented.
The Risks of Deepfakes
Deepfakes introduce additional risks because AI can generate realistic representations of people, including their faces and voices. These capabilities can make fabricated media appear credible, particularly when it imitates someone the audience already trusts.
Common risks include:
Identity impersonation: Synthetic faces or cloned voices can be used to impersonate employees, executives, public figures, or family members.
Financial fraud: Criminals may use fabricated audio or video to support urgent payment requests or manipulate financial decisions.
Non-consensual intimate imagery: Deepfake technology can create fabricated sexual images or videos of people without their consent.
Political misinformation and disinformation: Synthetic recordings can falsely portray public figures making statements or engaging in actions that never occurred.
Business and workplace deception: Manipulated video calls or voice messages may be used to bypass trust-based verification procedures.
Loss of public trust: As synthetic media becomes more convincing, people may become less confident in genuine recordings and authentic evidence.
The risks are not limited to highly sophisticated deepfakes. Easily accessible tools can also make certain forms of synthetic impersonation available to people with limited technical expertise.
These harms are not exclusive to one category. What differs is the mechanism: shallowfakes can exploit the apparent credibility of authentic recordings, whereas deepfakes can create statements or appearances that never existed. Because each may pass a quick visual check, the next section distinguishes the evidence needed to assess them.
How to Detect Deepfakes and Shallowfakes?
Detecting deepfakes and shallowfakes requires more than simply looking for unusual visual details. Deepfakes may contain AI-generated faces, voices, or movements, while shallowfakes often rely on selective editing or misleading context. Because both can appear convincing, the most reliable approach combines careful media inspection, source verification, contextual checks, and appropriate detection tools.

1. Check the Original Source
Start by tracing where the image, video, or audio recording was published and whether an original file can be identified. The earliest upload available online is not necessarily the recording’s original source. Likewise, an official-looking account or familiar publisher can be compromised or mistaken, so attribution should be evaluated alongside other evidence.
For shallowfakes, locating the original version can reveal whether footage has been cropped, edited, or presented without important context. For deepfakes, checking the source can help establish whether the purported speaker actually released the recording.
2. Look for Visual and Audio Inconsistencies
Some deepfakes may contain irregularities such as unnatural facial movements, inconsistent lighting, unusual blinking patterns, or imperfect synchronization between speech and lip movements. Cloned voices may also exhibit unusual pronunciation, rhythm, or changes in tone.
However, these signs are not conclusive. Compression, lighting, recording quality, and network delays can create similar effects in genuine media. Sophisticated synthetic media may show none of the expected artifacts. Treat visible or audible anomalies as investigative leads, not as proof of manipulation—and do not treat their absence as proof of authenticity.
3. Examine the Context and Editing
Shallowfakes often become misleading because genuine material has been edited or presented in the wrong context. Review the complete video or recording rather than relying on a short clip shared on social media.
Compare the clip with any available full-length recording and establish the event timeline. Verify where and when the depicted event happened, not merely when a post was uploaded. Check whether the caption names the correct people, place, and incident. A selectively edited speech can change a statement’s meaning, while an unaltered older video can be relabeled as current footage.
4. Use Deepfake Detection Tools Carefully
AI-based detection tools can analyze images, video, or audio for patterns associated with synthetic media. Depending on the tool, they may assess facial features, visual inconsistencies, voice characteristics, or other technical signals.
These tools have limitations. Performance varies with media quality, generation methods, and the type of content assessed. A detector’s estimate of possible AI generation is not the same as evidence about a file’s origin, nor does it establish whether a caption is factually correct. Most automated deepfake detectors cannot determine that genuine footage was selectively quoted. Treat results as one clue, not a final judgment.
5. Verify Suspicious Requests Independently
When manipulated media is used to support a financial request, identity claim, or urgent instruction, verify it through a separate trusted channel. For example, contact the person using a previously verified phone number rather than replying to the suspicious message or relying on the same video call.
This is particularly important in business settings, where a convincing voice or video may be used to imitate an executive, employee, or supplier.
6. Consult Independent Sources
Unlike tracing an upload’s origin, independent corroboration tests whether the underlying claim is true. Look for unrelated footage, contemporaneous records, direct statements, reputable reporting, or fact-checking work that independently supports or contradicts the account. Multiple reposts of the same clip do not count as independent confirmation.
For potentially harmful material, avoid forwarding it before its authenticity and context have been checked. Sharing a manipulated clip, even to question its accuracy, may increase its reach.
When the available evidence is incomplete, avoid treating an uncertain finding as a confirmed fake or confirmed authentic recording. Document the remaining uncertainty and seek further verification before repeating high-stakes allegations.
How to Prevent Deepfake and Shallowfake Misuse?
Fighting deepfakes and shallowfakes requires a combination of technology, public awareness, responsible media practices, and legal safeguards. Individuals and organizations can reduce the risks associated with manipulated content by following practical verification and prevention measures.
1. Establish Provenance and Media Controls
Organizations can reduce uncertainty about important media by protecting authoritative originals and documenting their handling. These controls make later verification more practical without claiming that technology can certify the truth of every statement.
Retain original recordings, associated creation details, and controlled copies for high-stakes communications where lawful and proportionate.
Establish consistent naming, access-control, and version-history practices so truncated clips are not mistaken for complete source recordings.
Publish contextual corrections and links to authoritative source materials through established organizational channels when misleading excerpts begin circulating.
Technical provenance controls support accountability, but their reliability depends on secure capture and handling throughout the media lifecycle. They complement, rather than replace, the investigative steps described earlier.
2. Improve Public Awareness and Media Literacy
Media literacy programs should teach the difference between an invented event and a genuine event presented misleadingly. Education is more effective when audiences practice assessing claims rather than memorizing a list of visual glitches.
Run training sessions that compare AI-synthesized audio with authentic recordings that have been cropped or miscaptioned.
Explain how emotionally charged headlines, misleading timestamps, and repetition can increase confidence in a false interpretation.
Encourage schools, publishers, and workplaces to discuss consent, satire, disclosure, and the responsible treatment of disputed evidence.
Evaluate training with realistic scenarios and constructive feedback, emphasizing careful reporting instead of public accusations based on uncertain clues.
This approach develops habits that transfer across formats and emerging manipulation methods. It also reduces the risk of dismissing legitimate evidence merely because sophisticated fabrications are possible.
3. Prevent Fraud and Impersonation
Manipulated media can make fraudulent requests appear genuine by imitating someone's face or voice. Strong identity-verification procedures can help organizations reduce this risk.
Verify unusual payment requests, confidential instructions, and sensitive business decisions through a separate, trusted communication channel.
Apply phishing-resistant multi-factor authentication to accounts and use separate identity and transaction-approval controls; MFA alone does not authenticate the words spoken in a call.
Establish clear approval procedures for financial transactions and changes to sensitive account information.
Train employees to recognize warning signs such as urgent requests, demands for secrecy, and unexpected changes to established procedures.
4. Address Legal and Ethical Concerns
Deceptive manipulation can raise privacy, harassment, impersonation, and reputational concerns. Ethical responsibilities—such as securing consent where appropriate, disclosing edits that materially change meaning, and correcting misleading publications—should be distinguished from legal duties, which depend on local law, the subject matter, and the circumstances. Not all ordinary edits or legitimate synthetic-media uses are unlawful or require identical disclosures.
Respect privacy and consent when creating, editing, or distributing content that depicts another person.
Disclose AI-generated or materially altered media when necessary to avoid misleading audiences.
Establish reporting procedures that allow people to flag harmful or deceptive content for review.
Preserve relevant original files, messages, links, and timestamps when fraud or harassment is suspected; screenshots can support an investigation but do not necessarily authenticate the original material.
Follow applicable laws and regulations, recognizing that legal requirements vary depending on the jurisdiction and circumstances.
Conclusion
The most important difference between deepfakes and shallowfakes is the nature of the manipulation. Deepfakes use AI to synthesize or substantially alter media, potentially making a person appear to say or do something that never happened. Shallowfakes generally rely on familiar edits or misleading context to change how authentic material is interpreted. Both can be used deceptively, but not every synthetic creation or ordinary edit is misleading.
A trustworthy assessment therefore looks beyond the surface of the content. Technical detection can inform questions about synthesis; provenance and contextual evidence address where a recording came from and what it actually depicts; independent corroboration tests the claims attached to it. For organizations, approval procedures, evidence retention, training, and responsible publication can reduce the potential harm.
The central takeaway is simple: neither a convincing appearance nor genuine underlying footage is enough, on its own, to establish that digital content tells the truth.





