Deepfakes: Do You Actually Understand Them — Or Have You Already Made Up Your Mind?
We all know what a deepfake is.
Or at least, we think we do.
Someone takes the face or voice of a real person, uses artificial intelligence to make that person apparently say or do something they never said or did, and releases it into the world.
Dangerous. Deceptive. Fake.
Case closed.
Except perhaps it isn’t.
A remarkable episode in Brazilian politics raises a much more difficult question for anyone working in social media, content creation, journalism, advertising or AI:
What happens when the deepfake tells you that it is a deepfake?
And if nobody is being deceived, should we still regard the technology itself as the problem?
The Bolsonaro deepfake
During a July 2026 political convention supporting Flávio Bolsonaro, an extraordinary figure appeared on a giant screen.
It was his father, former Brazilian president Jair Bolsonaro.
Except it wasn’t.
Bolsonaro’s image and voice had been recreated using artificial intelligence.
More importantly, the creators did not attempt to hide this.
The synthetic Bolsonaro effectively began by telling his audience:
What you are seeing isn’t really me. My image and voice have been simulated using artificial intelligence.
Only then did the avatar deliver its political message of support for Flávio Bolsonaro.
That distinction matters enormously.
This wasn’t an attempt to convince an unsuspecting audience that an artificial video was genuine.
The artificial nature of the content was part of the presentation.
And that turns what initially sounds like another frightening deepfake story into something much more complicated.
There are several different things we’re calling “deepfakes”
Consider four scenarios.
Level 1 — The malicious deepfake
Someone creates a realistic video of a politician, business leader, celebrity or ordinary person saying something they never said.
There is no disclosure.
The objective is for people to believe it is genuine.
Artificial content + impersonation + deception.
This is the deepfake we have learned to fear.
And with good reason.
Level 2 — The disclosed synthetic recreation
Now change one fact.
The video begins:
“This is an AI-generated recreation.”
Nobody is being asked to believe that the pixels they are watching represent a real recording.
Suddenly the ethical question changes.
The content is still synthetic.
The person still didn’t physically stand in front of that camera.
But where is the deception?
Level 3 — The authorised digital twin
Take it another step.
Imagine that I explicitly authorise an AI representation of myself.
It looks like me.
It sounds like me.
It might even be trained on years of my writing, videos and opinions.
Every piece of content prominently says:
AI-generated representation of Nick Godliman.
Could my digital version introduce a video?
Could it translate my videos into Portuguese while retaining my appearance and voice?
Could it deliver a product explanation?
Could a famous actor license their digital identity to an advertiser?
Could a CEO use an authorised digital twin to deliver routine corporate communications?
We’re already moving a long way from the conventional idea of a “deepfake”.
Level 4 — The person who is no longer there
Now things become even more interesting.
Imagine a filmmaker dies before completing a documentary.
Their family and estate authorise an AI recreation to read an introduction that the filmmaker actually wrote.
The programme clearly announces:
“The following sequence uses an authorised AI recreation.”
Nobody has been deceived.
Or imagine a historical documentary recreating Winston Churchill reading words from an authenticated letter for which no recording exists.
Again, clearly labelled.
Is that unethical?
Or is it simply another filmmaking technique?
Brazil makes this particularly fascinating
The Bolsonaro example contains another unusual element.
The real Jair Bolsonaro was subject to legal restrictions affecting his ability to communicate publicly.
Yet an artificial Bolsonaro appeared.
That raises a question that legislators, courts, platforms and creators are going to have to confront:
If a person cannot speak, can their digital representation speak for them?
And who, legally, is actually speaking?
The person whose likeness is being used?
The person who created the AI?
The organisation publishing it?
The software?
These questions extend far beyond Brazilian politics.
They potentially affect creators, influencers, actors, musicians, businesses and anyone whose identity has commercial value.
Brazil’s court had to confront exactly this problem
The Brazilian electoral authorities did not simply conclude:
“It was labelled, therefore it was legal.”
Brazil has introduced strict rules governing synthetic media and electoral propaganda.
The Bolsonaro case eventually reached Brazil’s Superior Electoral Court.
In September 2026, the court decided by 4–3 not to punish Flávio Bolsonaro over this particular AI appearance.
The distinction was important.
The court considered the circumstances of the party convention and the nature of the message, including the distinction between asking for political support and explicit electoral propaganda asking for votes.
So disclosure mattered.
But disclosure alone was not a magic legal defence.
That distinction is important for creators everywhere.
There is another problem: the label can disappear
Imagine I publish a 60-second AI video.
The first five seconds say:
THIS VIDEO CONTAINS AN AI-GENERATED RECREATION.
Someone downloads it.
They remove the first ten seconds.
They upload the remaining 50 seconds to TikTok, Instagram, X, Facebook or YouTube.
My disclosed synthetic video has suddenly become an undisclosed deepfake.
I didn’t create that deception.
But the deception now exists.
That suggests something much more sophisticated may eventually be necessary.
Perhaps disclosure shouldn’t merely appear before synthetic content.
Perhaps it must travel with it.
Persistent onscreen identification.
Digital watermarking.
Embedded provenance.
Content Credentials.
Metadata capable of identifying AI generation.
Systems designed to survive copying, clipping and reposting.
That could become one of the most important technical and regulatory questions facing social platforms.
The real dividing line may not be “real versus fake”
For years, our instinct has been:
Real = good.
Fake = bad.
AI makes that increasingly inadequate.
Perhaps the more useful distinctions are:
Disclosed or undisclosed?
Authorised or unauthorised?
Accurate representation or fabricated claim?
Provenance preserved or provenance removed?
Designed to inform or designed to deceive?
Those are much harder questions.
But they may also be much better ones.
Consider your own digital identity
For content creators, this isn’t theoretical.
Imagine an AI model trained on everything you have ever publicly created.
Your videos.
Your articles.
Your podcasts.
Your voice.
Your expressions.
Your opinions.
Your knowledge.
At some point, an extraordinarily convincing digital version of you becomes technically possible.
The question then isn’t simply:
“Can somebody deepfake me?”
It becomes:
“Under what circumstances would I allow an AI version of me to exist?”
Could you license it?
Could your company use it?
Could somebody else own it?
Could your family inherit it?
Could it continue producing material after your death?
Could you revoke permission?
And how would an audience distinguish between an authorised digital you and somebody else’s counterfeit version?
These aren’t questions for some distant science-fiction future.
They are becoming questions about identity, intellectual property, consent and content creation now.
Perhaps we’ve been using the wrong word
“Deepfake” has become almost entirely pejorative.
Understandably so.
But we may eventually need much better language.
A malicious impersonation designed to fool voters and an authorised, clearly labelled AI recreation are technologically related.
Ethically, they can be worlds apart.
Calling both simply “deepfakes” risks confusing the technology with its use.
Photography can document a war crime or fabricate evidence.
Photoshop can correct exposure or falsify a photograph.
Video editing can tell a story or deliberately misrepresent one.
AI-generated human likenesses may ultimately require the same distinction.
So here’s the challenge
Before automatically condemning the next “deepfake” you encounter, ask five questions:
- Was the use of AI disclosed?
- Did the person represented consent to its use?
- Does the synthetic person communicate something they actually authorised?
- Could a reasonable viewer mistake it for genuine footage?
- Will the disclosure remain attached when the content is copied or edited?
The answers may tell us far more than the word “deepfake” ever could.
Brazil has just provided an unusually powerful example.
Whatever your opinion of Bolsonaro — favourable, hostile or completely indifferent — isn’t really the point.
The technology is.
A synthetic politician appeared before an audience and effectively announced:
I am artificial.
That leaves us with a question every serious content creator should probably start considering:
If everybody knows the fake is fake, is the “fake” itself really the problem?
Or was deception the problem all along?
S.M.A.C.C. — Social Media and Content Creators Network
Examining the technology, law and responsibility behind modern content creation.
