Skip to content
back
ART-03·CRITIQUE·8 January 2026·9 min

The truth crisis: what you see may no longer be real

By Camilo Rivera

Screenprint poster: two identical discs, one yellow, one violet and slightly out of register, like an original and its copy
Generated poster · art direction by OSOM Labs

A video of a president declaring war. A photo of an explosion at the Pentagon. A voice message from your boss asking for an urgent bank transfer. All of this existed. All of it was fake. In 2026, the question is no longer "is it true?" but "how do you know if it's true?"

0101

What has changed

A few years ago, creating a fake video required Hollywood-level production. Today, with a laptop and a few tools anyone can use, you can:

  • Generate a photorealistic image in seconds
  • Create a convincing voice imitation with just a few seconds of audio and an accessible tool
  • Produce a video of a person saying anything
  • Generate thousands of "news" articles automatically
  • The problem: Our brains haven't evolved. We continue to believe what we see. And what we see can now be entirely fabricated.

    March 2023: An image of Pope Francis in a Balenciaga puffer jacket goes viral. Many people believe it. It was a Midjourney image generated in minutes. (Source: AFP Factuel, BuzzFeed News)

    May 2023: A fake photo of an explosion at the Pentagon causes the American stock market to dip for several minutes, for an image generated in a few clicks. (Source: Reuters, Bloomberg)

    0202

    The 4 types of fakes we encounter daily

    1. Generated images

    Midjourney, DALL-E, Stable Diffusion produce images indistinguishable from real photos. Landscapes, portraits, news scenes: anything is possible.

    2. Video deepfakes

    One person's face on another's body. Or worse: an entirely synthetic person speaking. "Fake CEO" video call scams already exist.

    3. Voice cloning

    With just a few seconds of your voice (a WhatsApp voice message can suffice), tools like ElevenLabs or Resemble.AI can generate a convincing imitation. The FBI warned about it in December 2024: criminals generate short audio clips with a loved one's voice to ask for urgent financial help or demand a ransom. Quality varies, but the element of surprise makes detection difficult in the heat of the moment.

    4. Automated text content

    Content farms generate thousands of "news" articles per day. Technically correct, but often false or misleading. Impossible to distinguish from a real article at first glance.

    0303

    How to detect fakes: the reflexes to develop

    For images:

  • Search for the source: A viral image with no source = suspicious. Do a reverse image search (Google Images, TinEye). Finding the original context remains the most reliable method, according to the experts consulted by AFP.
  • Look at the details: Odd hands, unreadable text, inconsistent backgrounds. Generators produce fewer and fewer of them: as early as March 2023, AFP warned that visual clues could not be relied on in the long run. A flaw is a clue; its absence proves nothing.
  • Be wary of perfection: A photo that's "too perfect" is often generated.
  • Don't rely on metadata: The date or camera recorded in a file tells you something when it's there. But social networks systematically strip it (AFP, 2023), and the C2PA, which publishes the image provenance standard, acknowledges that it can be removed. Its absence proves nothing.
  • For videos:

  • Watch the eyes: Irregular blinking, fixed gaze, inconsistent reflections.
  • Listen to the voice: Strange rhythm, flat intonation, missing breathing.
  • Search for the original video: If someone "says" something shocking, look for other sources.
  • For personal messages (voice, video):

  • Call the person back: If your "boss" asks for an urgent transfer via voice message, call them directly.
  • Ask a personal question: Something only the real person would know.
  • Don't give in to urgency: Time pressure is the main weapon of scams.
  • 0404

    The psychological traps

    Fakes don't work through technology alone. They exploit our cognitive biases, mental shortcuts we all use:

    Confirmation bias

    We more readily believe what confirms what we already think. Information that proves us right slips through unchecked.

    The repetition effect (illusion of truth)

    The more we see something, the more we believe it, even if we know it's false. This is documented in cognitive psychology. Content farms exploit this: repeating a lie across dozens of different sites makes it "true" in our memory.

    Apparent authority

    A verified account, a site that looks like a known media outlet, a "Dr" before a name. The appearance of authority disables our critical thinking.

    Emotion as a short-circuit

    When we're angry, scared or outraged, we react before we think. Viral content is optimised for this: triggering a strong emotion BEFORE we think. If a piece of news makes you furious, that's the moment to slow down.

    An MIT study published in Science in March 2018 measured it on Twitter: among the news stories checked by fact-checking organisations that circulated there from 2006 to 2017, falsehood travelled further and faster than the truth. True stories took about six times as long to reach 1,500 people. The researchers link the gap to the novelty of false news and to the emotions it provokes: fear, disgust and surprise. And the gap didn't come from bots: it came from humans.

    0505

    What we can do in practice

    In daily life:

  • Wait before sharing. Virality rewards speed, not truth.
  • Diversify your sources. If you only read one type of media, you're vulnerable.
  • Accept doubt. "I don't know if it's true" is a valid response.
  • To protect yourself personally:

  • Limit videos/audio of yourself online (less material for cloning).
  • Establish a "family password" for emergencies: if someone calls in distress, ask for the word. The FBI recommends the same.
  • Be wary of urgent money requests, regardless of the apparent source.
  • For society:

  • Educate those around you. Many people don't know these technologies exist.
  • Don't share fakes "to show they're fake": you amplify their reach.
  • Support quality journalism. Verification is expensive. Lies cost next to nothing.
  • 0606

    Our position

    We explore AI every day. We see what it can do: the best and the worst.

    What we think:

    Technology isn't the problem. It's a tool. The problem is that we haven't adapted our reflexes to this new reality.

    For millennia, "seeing was believing". That rule is obsolete. We need to relearn how to doubt, not everything, but what hasn't been verified.

    The good news: Once you know that fakes exist, you become much harder to fool. Awareness is the first defence.

    Key takeaways

    01

    Images, videos, voices: everything can be convincingly generated in 2026.

    02

    Reverse image search is your best friend.

    03

    An odd detail is a clue; its absence proves nothing, and neither does missing metadata.

    04

    Urgency and strong emotion are warning signs.

    05

    "I don't know if it's true" is an intelligent response, not an admission of weakness.

    06

    Awareness of the problem is already a form of protection.

    Stack

    DeepfakesVoice CloningGenerative AIFact-checking
    deepfake detection · fake news · voice cloning · ai images · disinformation · verify information · truth crisis

    More articles in the journal

    We document everything we do.