We live in an era in which artificial intelligence, once a science fiction topic, now shapes deep aspects of society – including how we understand and trust reality. One of the adverse reactions to this technological advancement is deepfakes: images, videos, and audios manipulated by generative AI, capable of mimicking faces, voices, and gestures with an extremely high level of realism.

The democratization of these tools, which are increasingly intuitive and easy to use, represents a turning point. Although the technology has creative and beneficial applications, its use for disinformation purposes has raised serious concerns in areas such as information security, politics, journalism, individual privacy, and even law. According to a Europol report, the use of deepfakes for malicious purposes is expected to increase exponentially worldwide, impacting not only elections and financial markets, but collective trust in institutions and information itself. How can we, as a society and technology professionals, prepare for this scenario where the truth can be fabricated in a few seconds?  

What are deepfakes – and why are they so concerning?

Deepfakes are synthetic media generated or altered with the help of deep learning algorithms, such as Generative Adversarial Networks (GANs). Basically, a GAN works with two neural networks in competition: a generator, responsible for creating the fake media, and a discriminator, which attempts to identify the fake content. Up to this point, the proportion of problem and solution seems to be the same, but the big catch is that these networks train each other, and the result of this process is increasingly perfect counterfeits.

At first, this innovation opened doors to very valuable artistic and educational creations. Examples include the recreation of historical figures in museums, the development of accessibility tools for people with speech impairments, and even the automatic dubbing of movies with perfect lip-syncing in multiple languages.

However, the same technology that enables these advancements feeds a catalog of malicious uses, and that is where concern with these technologies comes into play. In addition to increasingly sophisticated fraud and extortion, deepfakes are the starting point for creating false political speeches, damaging the reputation of individuals and companies, and non-consensual pornography, in which images of individuals, mostly women, are placed into pornographic content, causing devastating psychological and reputational damage. In cases involving politics, the effects of these innovations can also be devastating, as this content shapes public opinion and can even be decisive in elections. 

The real risks of deepfakes for information security

In the context of information security, deepfakes represent a multi-faceted threat. Cybercriminals are already using fake audio to simulate the voices of CEOs and carry out financial fraud. The 2019 case, in which a British energy company lost $243,000, was just the beginning. In a more recent and impactful incident in 2024, an employee of a multinational in Hong Kong was tricked into transferring $25 million after participating in a video conference with digital recreations of his colleagues, including the company's CFO.

In addition to direct financial fraud, the risks include using audio or video to request access to critical systems or sensitive information, as well as spreading fraudulent links in content that appears to come from trusted sources. Another real risk of this technology is generating panic through fake news about the financial market or widely used payment methods, such as Pix.

And when distrust begins to reign?

To increase the level of complexity even further, this issue has made it possible to deny the truth, dismissing it as if it were a deepfake. This phenomenon is increasingly common and even has a name: the Liar's Dividend (Liar's Dividend). In this context, the social impact of deepfakes goes far beyond a simple distinction between true or false, characterized by suspicion toward any and all information.

A politician caught in a compromising situation, for example, can simply claim that the content is a hoax generated by AI, creating enough doubt to escape accountability. 

The human brain, programmed to trust visual and auditory cues, is particularly vulnerable. A well-executed fake video generates a much stronger emotional response than text, making it more persuasive and likely to go viral before any fact-checking can take place. If everything can be faked, who can be trusted? The weakening of a common baseline of facts is fertile ground for polarization, extremism, and large-scale political manipulation.

How to identify a deepfake: signs, tools, and habits

Although technology is advancing rapidly, many deepfakes still show flaws and imperfections. Awareness of this and an attentive eye are the first defense mechanisms. When coming across doubtful content, observe:

• Blinking and eye movements: analyze eye movements, such as very frequent or almost non-existent blinking, noting if they look natural.
  
  
  

• Lip-syncing: try to notice if there are slight misalignments between the sound of the voice and the movements of the mouth.
  
  
  

• Face and hair details: see if the outline of the face is blurry, if the skin texture is excessively smooth, or if strands of hair look artificial, with unusual movement.
  
  
  

• Lighting and shadows: look for inconsistencies in form, such as the light on the face compared to the rest of the environment.
  
  
  

• Audio quality: robotic audio with strange intonation or synthetic background noise is usually noticeable. Listen carefully and try to identify if the content is fake.

In addition to observation, it is important to verify the source and even rely on some fact-checking website. Tools like Microsoft Video Authenticator and Intel FakeCatcher are being developed to help combat this type of content. But the long-term solution also lies in the habits of those who use the internet.

The role of critical thinking in this new era

More than any tool, the main antidote against disinformation is critical thinking and media literacy. In a world flooded with artificial content, the ability to question the origin of information is an essential skill.

• Pause before sharing: disinformation feeds on impulsivity and confirmation bias. Stop for a moment and ask yourself: "Does this information seem too good (or too bad) to be true?".
  
  
  

• Verify the source: does the content come from a reliable media outlet? Does the account that shared it have a solid history? And even if it has a good history, always consider any unusual behavior.
  
  
  

• Triangulate: is the same news being reported by other independent and credible sources?
  
  
  

• Use reverse search: tools like Google Lens and TinEye allow you to check if an image or a video frame has already been used in another context.

Deepfakes and legislation: are we prepared?

As expected, legislation moves at a slower pace than technology, but the issue is urgent; after all, the regulatory challenges are immense. How to punish malicious use without bumping into topics like false accusation? Should liability fall on whoever creates, whoever shares, or on the platforms that host the content? 

Finding this balance has been one of the central dilemmas of modern times, regardless of nationality. In the European Union, the AI Act (Artificial Intelligence Act) has already been approved and establishes transparency rules, requiring deepfakes to be clearly labeled as artificial content. In the United States, several states have progressed on the matter independently, but there is still no federal legislation for the AI industry. In Brazil, the debate takes place mainly within the scope of Bill 2630, known as the Fake News Bill, and in discussions about updating the Brazilian Civil Rights Framework for the Internet and the General Data Protection Law (LGPD) to address the risks of generative AI.

Paths to a more trusted future

Technology alone will not solve a problem that is fundamentally human and social. The paths to a more trusted future in this regard involve a multi-faceted and collaborative approach. 

Governments and legislators need to create agile and balanced regulatory frameworks that punish malicious use without stifling innovation and freedom. Companies, schools, and governments have the responsibility to invest in digital education and media literacy starting from basic education, empowering people not only to consume but to interpret and share information in an ethical and responsible way.

Technology platforms themselves must proactively invest in detection tools, clear labeling of artificial content, and transparency regarding their recommendation algorithms, which currently accelerate virality. 

The press and fact-checkers, more than ever, must be guardians of truth, adopting new verification tools, in addition to educating the public. As for us, citizens, it is our responsibility to cultivate healthy skepticism and become agents of containment rather than propagation of disinformation.

Truth will continue to be a socially negotiated value. In the landscape shaped by generative AI, our ability to navigate reality will depend less on the perfection of technology and more on ethical and collective discernment.