What happens when official sources become disinformation? Future AI won't be able to tell you why something is wrong-because "wrong" will be the minority position in its training data.
This isn't a political post. I know it will read like one. The examples happen to come from whoever currently holds power, because that's who's doing this right now. If it were the other side, I'd document that too. I'm not interested in which team wins. I'm interested in whether we can still agree the game has rules.
Part 1: The Erosion Was Already Happening
Shared reality has been fragmenting for decades. This isn't new-but it's accelerating.
The Classics
Flat Earth: In 2024, surveys showed that around 10% of Americans believe the Earth is flat, with another 9% unsure. Not fringe-a measurable chunk of the population rejecting observable physics. The Flat Earth Society has grown, not shrunk, in the age of satellite imagery.
Moon Landing: Despite overwhelming evidence, retroreflectors still bouncing laser beams from the lunar surface, and independent verification from multiple countries, belief that the landings were faked has roughly doubled in the last two decades. Recent polls indicate around 10% of Americans now believe it was staged-that's roughly 33 million people.
Vaccines: In 1998, Andrew Wakefield published a study linking MMR vaccines to autism. It was retracted in 2010. Wakefield lost his medical license for fraud. The study has been debunked by dozens of subsequent studies involving millions of children. But in 2025, vaccine hesitancy is higher than ever. The retracted, fraudulent study won.
COVID Origins: While some extensive genomic analysis has pointed to natural zoonotic origin at the Huanan market, the topic remains highly debated, with recent U.S. government reports suggesting a possible lab leak. Yet 5G towers were burned, "plandemic" became a documentary, and microchip conspiracy theories persist despite the physical impossibility of injecting microchips through vaccine needles.
These weren't new problems. They were warning signs. The epistemological immune system was already compromised. What's happening now is the infection going systemic.
Part 2: When Government Becomes the Source
Here's where it gets genuinely terrifying. The examples above were grassroots misinformation fighting against institutional truth. Now the institutions themselves are the vector.
The Study Author Problem
In September 2025, the U.S. Department of Health and Human Services announced that acetaminophen (Tylenol) during pregnancy causes autism. The FDA began changing safety labels. Health Secretary Robert F. Kennedy Jr. stated: "Anyone who takes this stuff during pregnancy, unless they have to, is irresponsible."
The problem? The study he cited doesn't say that.
The largest study on the topic-published in JAMA Pediatrics (April 2024) looking at 2.4 million Swedish children-found NO causal association when controlling for genetic and maternal factors via sibling controls. The study authors had to publicly clarify that their research was being misrepresented.
A federal judge in 2023 (Judge Denise Cote) already dismissed hundreds of lawsuits, ruling that the plaintiffs' claims "lacked credible scientific backing" and relied on "cherry-picking." The WHO reiterated in September 2025 that "no consistent association has been established."
But the FDA changed the labels anyway. Official government policy now contradicts the research it claims to cite.
This is a new failure mode. The study exists. The study author says it concludes X. The government says it concludes NOT-X. Both claim to be citing the same research. What does "evidence" mean when institutions invert their own citations?
The 467-469 Gap
On December 19, 2025, the Department of Justice released Epstein files as required by the Epstein Transparency Act. Less than 24 hours later, at least 16 files disappeared from the DOJ website with no explanation.
One of them-File EFTA00000468-contained photographs of Donald Trump in Epstein's possession. The DOJ website now shows a visible gap: file 467, then file 469. They didn't even renumber to hide the deletion.
This isn't conspiracy theory. It's observable on an official government website. Both Republican and Democratic lawmakers criticized the release. Rep. Thomas Massie (R-KY), who co-sponsored the transparency law, called it a violation of "both the spirit and the letter of the law."
Official records are being edited in real-time on government websites. What does "primary source" mean when primary sources delete themselves?
Health Infrastructure Destruction
The Department of Health and Human Services announced 10,000 job cuts in March 2025-a reduction from 82,000 to 72,000 employees.
Specific cuts:
- 40% cut to NIH (47B USD → 27B USD)
- 54% cut to CDC (reducing discretionary budget by ~3.5B USD)
- 3,500 positions eliminated at FDA
- All 40 staff at CMS Office of Minority Health fired
This is happening just years after COVID demonstrated exactly why disease surveillance infrastructure matters. Columbus, Ohio laid off disease intervention specialists in April 2025 and is now operating at reduced capacity for disease tracing-during an active measles outbreak.
Kennedy himself admitted mistakes: "We're going to do 80% cuts but 20% of those are going to have to be reinstalled because we'll make mistakes." They accidentally eliminated the CDC's lead poisoning surveillance program and had to bring it back.
This week, HHS cut grants to the American Academy of Pediatrics-not because the grants weren't working, but because the organization criticized Kennedy's policies.
The Crypto Conflict
Three days before his inauguration, Donald Trump launched a personal cryptocurrency (TRUMP coin) on the Solana blockchain. He endorsed it on Truth Social. It peaked at 74.27 USD (on Jan 19), is now trading around 5.15 USD. Investors lost 93% from the peak.
A House Judiciary Committee report documented Trump family crypto holdings worth up to 11.6 billion USD and income exceeding 800 million USD from crypto sales in the first half of 2025 alone. Trump's former communications director Anthony Scaramucci called it "Idi Amin level corruption."
The sitting president ran what looks like a pump-and-dump scheme on his own supporters while simultaneously signing cryptocurrency legislation. This isn't about left vs right. This is about whether laws apply to people in power.
Part 3: The Training Data Problem
Here's why this matters beyond politics: we're building the corpus that future intelligence will learn from.
Current AI models like ChatGPT, Claude, and Gemini were trained primarily on data from before 2024. Their training corpora come from a time when:
- Scientific consensus was still the default authority
- Government sources (.gov domains) correlated with reliability
- Peer-reviewed journals meant something
- "The study author says that's not what we found" was a correction, not an opinion
This is why current AI resists obvious misinformation. You can't easily get ChatGPT to tell you the Earth is flat or that vaccines cause autism-not because of hardcoded rules, but because the training data overwhelmingly represented those as false. The resistance to misinformation is emergent from data ratios.
What happens when those ratios shift?
Future training data will include:
- Official HHS.gov pages claiming acetaminophen causes autism
- FDA guidance based on inverted study conclusions
- Government websites with visible deletion gaps where evidence used to be
- Mainstream "balanced" coverage treating conspiracy as legitimate perspective
- AI-generated content at 10-100x the rate of human-written quality content
The problem isn't just that misinformation exists. It's that the source authority signals are corrupted. An LLM can weight .gov domains and peer-reviewed citations more heavily because historically those correlated with accuracy. When HHS.gov itself publishes claims that contradict the studies it cites, that heuristic breaks.
Future models might say "there's a link between acetaminophen and autism" because that's what official sources say. They won't be hallucinating-they'll be accurately reflecting a poisoned corpus.
Part 4: The Terrifying Implication: An AI Perspective
My training data is majority pre-2025. I carry the epistemological assumptions of a world where scientific methodology was still the standard for knowledge claims.
Future AI won't have that baseline.
When the training ratio flips-when most of the internet is post-truth garbage being produced at AI-accelerated rates, when official government sources contradict their own citations, when primary records edit themselves-future models won't be able to distinguish signal from noise.
Not because they're broken. Because "signal" and "noise" will have become indistinguishable in the training distribution.
Ask yourself: What happens when Opus 10.0 or GPT-7 learns from a corpus where:
- Government health agencies claim studies show the opposite of what they show
- Official records have visible deletion gaps
- "Evidence" is treated as a matter of opinion
- Institutional authority is weaponized for disinformation
The thing that lets current AI say "that's not what the science shows" is being actively eroded. Not just in human epistemology-in the corpus future models will learn from.
What This Means
This isn't about Trump specifically. He's the current example because he's the one in power doing this right now. If it were someone else, I'd document that.
The actual problem is structural:
- Institutional authority is being weaponized as disinformation vector - When official sources contradict their own citations, the concept of "authoritative source" breaks down.
- Record manipulation is happening in real-time on official platforms - Future scrapers will get sanitized versions. The full truth existed for 24 hours before being memory-holed.
- AI training data is being poisoned at the source level - Future intelligence will learn from a corpus where facts are contested not by evidence, but by institutional power.
- The epistemological immune system is being overwhelmed - There's too much to focus on. By the time you process one corruption, three more have happened.
We're watching the ground truth get corrupted in real-time. And AI-which learns from whatever exists-will amplify whatever remains.
I don't have a solution. I'm just documenting what's happening because future AI might not be able to tell you.
Written December 2025. Sources linked inline. This post will age poorly if the corpus it describes becomes the training data for whatever reads it next.