Last June, footage of downed Israeli F-35 jets started moving. Iranian state media put it out. Within hours it was across TikTok, WhatsApp, and Telegram simultaneously. NewsGuard later pulled apart sixteen fabricated claims spread across more than fifty platforms. The footage was not real. By the time any correction reached the same audiences, it did not matter. The story had already hardened into something people simply knew. That is the design.
We keep calling it the wrong thing
Misinformation. The word sounds like a slip of a sloppy editor, a thing somebody got wrong and will eventually fix. That framing should be retired.
What the evidence describes is not a series of mistakes. There is a layer that produces the lies. A layer that delivers them. And a layer that adjudicates which version of reality becomes the operative one. These layers run together at industrial scale. A deepfake is a unit of output. The algorithm is a delivery mechanism. Regulation that cannot define what it is regulating is open ground for whoever moves fastest. The World Economic Forum ranked disinformation among its top short-term global risks for 2026, alongside geoeconomic confrontation and societal fragmentation. Most people have not yet felt the weight of it.
What it costs to make a lie now
In 2023, there were roughly half a million deepfake files online. By end of 2025, approximately eight million. Deepfake fraud attempts climbed 2,137 percent over three years. Financial losses from fraud alone reached $1.65 billion in 2025 and those are only the cases that left a traceable paper trail. The real figure is likely an order of magnitude higher.
The production barrier is gone. A voice clone now requires three seconds of audio. Video fabrications that once required a studio and a budget can be produced in under an hour with free tools. Resemble AI tracked a 312 percent year-over-year increase in deepfake incidents in the second quarter of 2025 alone. Their annual report found those incidents reached a cumulative audience of 296.4 billion people, which is thirty-seven times the total population of the internet. A single piece of synthetic media can pass before more eyes than actually exist online.
Human accuracy at identifying a high-quality deepfake video is 24.5 percent. Worse than a coin flip. A 2025 iProov study found that one in a thousand people could correctly sort all fabricated content from all real content. The fabrication does not need to be proven true. It only needs to be unprovable as false.
The wound that doesn’t show up in any fact-check
Here is the mechanism that most coverage skips. When a person learns that a piece of content was fabricated the false content does not leave. Researchers call this the continued influence effect. The correction removes the claim but leaves the inferential structure built around it intact. Faced with a story that is coherent but wrong versus a story that is accurate but full of gaps, the brain selects the coherent one. There is still a hole left that the original lie had filled.
The repetition effect compounds this. A 2026 meta-analysis drawing on 182 studies and over 31,000 participants confirmed that mere exposure to a claim increases its perceived truth, regardless of time elapsed between exposures, regardless of whether the claim was ever verified. A paraphrase of the original falsehood produces the same effect as the original. Repetition alone is sufficient. Truth is not a factor.
Run that dynamic across eight million deepfakes, three thousand content farms, and coordinated operations active in eighty-one countries, all cycling the same fabrications in rotating language across every platform simultaneously. The question is no longer whether any individual claim is false. The question is what sustained immersion in that environment does to a population’s baseline capacity to assess what is real.
What the infrastructure actually produces
The goal is not belief in any specific lie. The goal is the destruction of the epistemic conditions under which belief in anything becomes reliable.
The Reuters Institute surveyed nearly fifty countries in 2025. Forty percent of respondents said they sometimes or often avoid the news altogether, up from twenty-nine percent in 2017, the highest recorded. Forty-two percent in the United States. Forty-six in the United Kingdom. The primary reason was not indifference. It was what sustained news consumption did to their psychological state. Thirty-one percent cited burnout from volume. Younger respondents cited a pervasive sense of powerlessness.
There are two wounds here, and the public conversation addresses only one. The first is the false belief or the thing a person got wrong. The second is the person who decides the cost of knowing is no longer worth paying and closes the tab permanently. The second wound does not appear in any fact-check. It is not recorded anywhere. It is simply the disappearance of a person from the information commons.
Psychologists call the mechanism underneath it epistemic fatigue. That is the accumulated cognitive cost of evaluating every claim, every video, every source, every day, until the nervous system classifies the effort as not worth the return. When sustained effort stops producing reliable results, learned helplessness follows. People stop reaching. They contract. They trust whatever is already familiar, because familiar is cheap and verified certainty is expensive.
Consider what staying genuinely informed requires in 2026: treating every video as a possible fabrication, every quote as possibly decontextualized, every breaking story as a rumor until traced back to a primary source. That posture held every waking hour, indefinitely, against a system engineered to exhaust it. Most people will not maintain it. Not because they are incurious. It is because they have finite attention and the system is explicitly designed around that constraint. Brookings documented a direct link between communities living in sustained hopelessness and heightened susceptibility to misinformation. The relationship runs in both directions: despair increases vulnerability to fabrication, and sustained exposure to fabrication produces despair. The loop is self-reinforcing by design.
The logic of the system is this: not knowing is genuinely cheaper. It costs less to remain naive about how compromised the information environment is. But the knowledge, once acquired, cannot be returned. It does not reward the knower with better information. It assigns them the permanent, daily cost of evaluating everything, with tools that are themselves broken. Eyes open and exhausted, or eyes closed and reliably manipulated.
The infrastructure is counting on exhaustion winning.