Artificial Intelligence · Deepfakes · Digital Communication
AI-generated deepfakes have turned a seemingly simple question into an increasingly complex problem: can we trust what we see and hear on a screen? The ability to recreate realistic faces, voices, and situations opens up enormous creative possibilities, but also new forms of fraud, impersonation, and disinformation.
For a long time, a photograph, audio recording, or video served as intuitive evidence that something had actually happened.
Seeing someone speak certain words seemed sufficient to assume they had actually said them.
Generative artificial intelligence is changing that relationship.
Today, it is possible to create synthetic images, voices, and videos capable of depicting people and events that never existed in the way they appear on screen.
The problem is not the existence of AI-generated content. The technology has legitimate applications in audiovisual production, education, entertainment, accessibility, and communication.
The risk arises when synthetic content is deliberately presented as authentic in order to deceive the person receiving it.
What Is a Deepfake?
A deepfake is synthetic or manipulated audiovisual content created using artificial intelligence techniques to convincingly depict a person saying, doing, or appearing to do something that did not happen that way.
It can involve:
- digitally generated or replaced faces;
- voice cloning;
- artificial lip synchronization;
- altered expressions and movements;
- fully synthetic images;
- AI-manipulated or AI-generated videos.
The term is commonly associated with fake videos of well-known people, but the problem is broader.
A scam can use a cloned voice, a manipulated video call, or an entirely fictional identity without ever producing the typical viral video of a celebrity.
Why Do Deepfakes Represent a Trust Problem?
The risk posed by deepfakes does not depend solely on someone believing a fake video.
There is a deeper consequence: uncertainty about whether any audiovisual content can be considered authentic.
As convincing synthetic content becomes easier to produce, images, audio, and video can no longer serve on their own as sufficient evidence.
This affects the way we interpret information related to people, companies, institutions, news, and public events.
The problem is not only believing something false. It is also beginning to distrust what is actually true.
Deepfakes Are Already Being Used in Real-World Fraud
The threat is no longer merely hypothetical.
Organizations such as the FBI have warned about the use of generative artificial intelligence to produce fake images, voices, and videos designed to facilitate different forms of fraud.
Criminals can use synthetic content to impersonate:
- company executives;
- authorities;
- celebrities;
- institutions;
- potential romantic partners or personal contacts;
- people known to the victim.
The objective may be to obtain money, information, login credentials, or convince the victim to take a particular action.
AI Can Make Impersonation More Convincing
Traditional social engineering depends on persuading someone to trust the person attempting to deceive them.
Artificial intelligence provides new tools for creating that appearance of legitimacy.
A voice may sound like a family member. A video may appear to show an executive. A video call can alter the appearance of the person on the other side in real time.
This means that some signals that once served as intuitive verification mechanisms are becoming less reliable.
Hearing a familiar voice should no longer be enough to authorize a significant financial transfer.
Seeing a familiar face during a video call should not replace authentication procedures when a sensitive request is involved.
The Economic Impact Can Already Be Measured
Fraud reports are beginning to reveal the economic scale of the problem.
The FBI Internet Crime Complaint Center’s (IC3) 2025 annual report stated that reported investment scams with an artificial intelligence nexus resulted in more than USD 632 million in losses.
The same report describes the use of AI-generated videos and voices of celebrities, CEOs, and other trusted figures to present fraudulent investment opportunities.
This figure does not represent all fraud involving deepfakes, nor does it allow every dollar to be attributed exclusively to fake videos.
But it does demonstrate that generative AI is already part of fraudulent operations with real economic consequences.
Deepfakes, Disinformation & Manipulation
Financial fraud is only one of the possible malicious applications.
Synthetic content can also be used to attribute false statements, fabricate events, or alter public perception of a person or institution.
There is also a second problem: once society knows that deepfakes exist, someone can attempt to discredit authentic content simply by claiming that it was generated using artificial intelligence.
The result is erosion in both directions:
- false content can appear to be true;
- authentic content can be presented as fake.
That is why the discussion around deepfakes should not be reduced solely to detecting imperfect images.
It is also a discussion about credibility, provenance, and digital trust.
How Can You Detect a Deepfake Video?
There are visual and auditory signals that may raise suspicion.
Among the indicators mentioned by the FBI are:
- unnatural movements;
- inconsistencies between the face and voice;
- unusual blinking;
- inconsistent shadows or lighting;
- abnormal body movements;
- unusual pauses or intonation;
- inconsistent background noise;
- audiovisual synchronization problems.
But there is one fundamental warning:
the absence of these flaws does not prove that a video is authentic.
Generative systems continue to improve, and some of the visual clues that once helped identify synthetic content are becoming increasingly difficult to detect.
We should not move from “seeing is believing” to “look for six fingers to know whether something is fake.” Verification requires more context.
How to Verify a Suspicious Video
When content could have significant consequences, it is safer to verify its provenance than to rely solely on guessing whether it was generated by AI.
1. Find the Original Source
Who published the video first? Does it come from the official account of the person or organization? Is there an original post that can be reviewed?
A cropped, forwarded, or downloaded copy loses some of the context needed to evaluate its provenance.
2. Look for Independent Confirmation
If the content shows an important event, check whether other reliable sources are reporting it.
An extraordinary video that appears only on an unknown account deserves greater caution.
3. Separate the Content From the Claim
Even an authentic video can be shared with a false description.
A real recording from another year, country, or event can be reused to construct a misleading narrative.
That is why it is not enough to ask whether the file was generated by AI.
We also need to verify whether it actually represents what the post claims it represents.
4. Verify Through Another Channel When Money or Sensitive Information Is Involved
If you receive a video call, audio message, or message from someone you know requesting money, passwords, codes, or confidential information, use another channel to confirm the request.
For example, contact that person directly using a previously known phone number instead of responding only through the channel from which the request arrived.
5. Do Not Act Based on Urgency Alone
Scams often attempt to reduce the amount of time available to think.
An urgent, secret, or extraordinary request should increase the need for verification, not reduce it.
Can a Tool Automatically Detect Every Deepfake?
There is no infallible solution capable of automatically determining the authenticity of every piece of content.
Detection tools can be part of the process, but their results need to be interpreted alongside other evidence.
This becomes especially important when content has been compressed, cropped, edited, re-recorded, or published multiple times across different platforms.
That is why digital verification is evolving from a strategy focused exclusively on detecting what is fake toward one that also seeks to demonstrate the provenance of what is authentic.
Content Credentials and C2PA: Verifying Where Content Comes From
One of the most important initiatives in this direction is C2PA, an open standard for incorporating verifiable information about the origin and history of digital content.
Las llamadas Content Credentials can provide information about how specific content was created or modified and which tools were involved in that process.
The goal is to provide provenance signals that can accompany images, audio, and video through different digital workflows.
This does not mean that a credential automatically determines whether a claim is true.
A file can be authentic and still be used out of context. Legitimate content may also exist without credentials.
Provenance is an additional signal within the verification process, not a substitute for judgment.
Digital Authenticity Requires Multiple Layers
The evolution of deepfakes demonstrates that there is unlikely to be a single technology capable of solving the problem of trust.
A combination of the following will be necessary:
- verifiable provenance;
- technical standards;
- detection tools;
- platform moderation;
- media literacy;
- journalistic verification;
- internal company protocols;
- human judgment.
The more realistic AI-generated content becomes, the more important it will be to have mechanisms that help establish where it came from.
What Should Companies Do About Deepfakes?
For an organization, the problem is not only reputational.
Impersonation can use the identity of executives, employees, or the brand itself to attempt to deceive customers and collaborators.
That is why it is advisable to establish procedures that do not depend exclusively on recognizing a face or voice.
Reasonable measures include:
- using multi-factor authentication;
- establishing procedures for approving payments and sensitive changes;
- confirming unusual requests through independent channels;
- training staff on social engineering and synthetic content;
- maintaining clearly identifiable official channels;
- defining how to respond to public impersonation of the brand.
Defending against deepfakes is both a technological and communication challenge.
What Responsibility Do People Who Create AI Content Have?
The existence of risks does not make AI-generated audiovisual content inherently negative.
These tools can be legitimately used to create characters, advertising, education, entertainment, prototypes, and new forms of audiovisual production.
The fundamental difference lies in intent and transparency.
Using AI to create a fictional scene that is clearly identified as such is very different from using someone’s face or voice to falsely attribute a statement to them with the intention of deceiving others.
For creators, companies, and communication professionals, transparency about the use of synthetic content may progressively become part of the relationship of trust with their audiences.
The Real Challenge of Deepfakes Is Preserving Trust
The quality of AI-generated video will continue to improve.
This means our ability to recognize fake content through intuition alone will probably become increasingly insufficient.
The answer cannot be to automatically distrust everything that appears on a screen.
Nor can it be to believe something simply because it looks real.
We need to develop a digital culture in which verifying the source, understanding the context, and checking provenance become a natural part of consuming information.
In the age of artificial intelligence, the question is no longer simply “does it look real?” but “how do I know where it came from?”
Deepfakes represent a threat when they use our trust as a mechanism for deception. But they are also forcing us to build a more mature relationship with digital information: one in which credibility does not depend solely on how convincing an image, voice, or video may appear.