On the night of March 8, 1971, eight individuals broke into an FBI field office in Media, Pennsylvania. They chose this date intentionally, knowing that everyone in the country, including anyone who might have noticed the pried-open door, would be focused on the boxing match between Ali and Frazier. The group took every file in the office, sorted through the material, and began mailing significant documents to newspapers.

This is how the United States first learned about COINTELPRO—not through oversight or a leak from a conscience-stricken official, but through a well-timed burglary during a boxing match. The FBI never apprehended them, and the public only learned their names in 2014, by which time most of them were grandparents, and the statute of limitations had long expired.

I think about those eight individuals whenever someone claims that artificial intelligence is about to ruin our shared relationship with the truth. The underlying implication is that we once had a clear relationship with the truth, but that may not be the case.

Here are a few examples of misinformation, much of which comes from government documents:

• In August 1964, the public was informed that North Vietnamese boats had attacked American destroyers in the Gulf of Tonkin. In response, Congress passed a resolution. Ultimately, fifty-eight thousand Americans lost their lives under its authority. However, the second attack that triggered all of this did not actually occur. Robert J. Hanyok, a senior historian at the NSA, confirmed this in internal documentation. Following Freedom of Information Act requests, Hanyok’s research revealed that signals intelligence (SIGINT) regarding the disputed August 4, 1964, Gulf of Tonkin incident was deliberately skewed to convince Congress and the White House that a second North Vietnamese attack had taken place. In reality, the North Vietnamese ships were conducting salvage operations, and the U.S. destroyers were firing at radar phantoms and weather anomalies in the dark.

• For forty years, the U.S. Public Health Service misled 600 Black men in Macon County, Alabama, by claiming to treat them for “bad blood.” These men were lured with false promises of free medical exams, meals, and burial insurance, yet they were never informed that they had syphilis. Even when penicillin became the standard treatment, the Public Health Service worked to keep it away from them. This unethical practice continued through the administrations of Truman, Eisenhower, Kennedy, Johnson, and Nixon, finally ceasing in 1972 when Peter Buxtun, an American epidemiologist, could no longer bear it. An Associated Press reporter named Jean Heller published his findings, leading to an article in the Washington Star on July 25, 1972, which became front-page news in the New York Times the following day.

• The Atomic Energy Commission assured residents living downwind of the Nevada Test Site that fallout was nothing to worry about, despite its own measurements indicating otherwise. According to the Nuclear Museum, the years of active nuclear testing in the United States spanned from 1951 to 1992, during which a total of 928 nuclear tests were announced. These included 99 above-ground atmospheric detonations, mostly in the 1950s and early 1960s, as well as over 800 underground tests designed to minimize surface fallout. Rural ranchers, local tribes, and farming communities suffered as their traditional ways of life and food sources—like contaminated milk from grazing cows—were poisoned. Long-term health problems, including cancer, thyroid damage, increased rates of congenital disabilities, fetal malformations, and compromised immune systems, emerged among residents. The last nuclear test, known as Divider, took place on September 23, 1992, marking the end of underground detonations before the U.S. instituted a nuclear testing moratorium. Eventually, public pressure led to government-backed dose reconstruction studies and the establishment of the Radiation Exposure Compensation Act (RECA) to assist affected citizens.

• Initially, the U.S. Army informed the family of Corporal Pat Tillman and the public that he had been killed by enemy fire in Afghanistan on April 22, 2004. Weeks later, following his memorial service and the awarding of a Silver Star, the Army admitted that he had actually been killed by friendly fire from his own unit. They had known within days that it was friendly fire, but the family found out five weeks later.

• In March 2013, the Director of National Intelligence testified before a Senate committee and was asked directly whether the NSA collected data on millions of Americans. He answered no. However, three months later, it became clear that this was not true. Under Section 215 of the USA PATRIOT Act, the NSA gathered bulk telephony metadata—including call durations, locations, and numbers dialed—covering millions of Verizon and other telecom customers daily. This specific bulk collection program officially ended following the passage of the USA Freedom Act in 2015. Additionally, operating under Section 702 of the Foreign Intelligence Surveillance Act (FISA), the NSA collected online communications and digital content directly from major U.S. technology companies like Google, Apple, Microsoft, and Facebook.

These examples represent just a fraction of what’s out there: Guatemala, plutonium injections, Cambodia, Iran-Contra, Camp Lejeune. The common thread throughout these paragraphs is that, eventually, the truth comes to light. Journalists and newspapers remain essential in revealing these truths.

Not one of these deceptions was exposed by oversight committees. None were flagged, labeled, downranked, self-reported to the public or moderated; they were uncovered by unglamorous, labor-intensive methods, such as the man who photocopied seven thousand pages at his friend’s ad agency after hours, a committee with subpoena power, a reporter with an editor willing to publish the story, and a historian delving into archives to read what was actually there.

The record of American deceit is essentially the same as the record of American honesty. You can’t tell the story of the Tuskegee experiment without mentioning Buxtun and Heller. Similarly, you can’t discuss the Pentagon Papers without referring to Ellsberg, the New York Times, and a Supreme Court that said no to the government. We can list these betrayals with names, dates, and numbers because people are ultimately quite adept at uncovering the truth.

So when I hear that AI is going to flood us with plausible misinformation, making it difficult to discern what is real, my honest reaction is: the flood has already happened. It has always been there. Lying has never been costly. A press conference is free. Iraq’s weapons of mass destruction were established with no technology at all—just a podium and a room full of people willing to report it and then question its validity. The Tonkin claim didn’t need a sophisticated system; it just required a classification stamp.

What really prevents the dissemination of falsehoods is not a lack of lies, but rather the existence of individuals and institutions with the authority, resources, and legal power to compel the production of evidence. This includes subpoenas, discovery processes, the Freedom of Information Act (FOIA), inspectors general, and whistleblower protections that ensure telling the truth does not endanger your life. Newsrooms equipped with evidence and lawyers are also crucial.

None of this machinery cares about the source of the lie. A subpoena does not inquire whether a person or an AI model generated the information it is investigating; it simply demands the documents.

What truly worries me isn’t deepfakes. It’s a public that believes nothing. That is more concerning than one that believes too much. Both groups end up in the same place, selecting their facts based on their affiliations, but the second group at least still acknowledges that evidence exists. When people are convinced verification has become impossible, you haven’t made them vigilant; you’ve rendered them inert and given a free pass to anyone lying to them.

Each sensationalized article about how we can no longer trust our own eyes ultimately aids the next liar by creating a universal defense. Is this recording authentic? It might be AI-generated. Is this memo genuine? AI could create it. If the next whistleblower, like Daniel Ellsberg, walks into a newsroom with a hard drive, the first question anyone asks might be, “How can we know?” Fakes harm the truth, but blanket deniability harms it even more.

Americans have not historically responded to being lied to by simply retreating; instead, they get furious and take action. This has led to significant legislation such as the Privacy Act, institutional review boards, and the Radiation Exposure Compensation Act. Each of these measures was established because someone was caught being dishonest, and each made it harder for that kind of deceit to happen again. That instinct didn’t disappear because AI took over the information highway.

I’m not claiming that technology changes nothing. Every case mentioned relies on an artifact—something tangible, whose existence serves as the proof. For example, there was Ellsberg’s stack of papers, the media files in a mailbox, and the Atomic Exposure Control’s own measurements. When producing a convincing artifact becomes nearly trivial while verifying one stays costly, the balance that supports the entire system starts to tip.

So, let’s tip it back. We need to implement provenance standards for recordings and documents. Invest real money in forensic capabilities. Give independent researchers real access to these systems, not just press releases about them. Establish liability for whoever deploys the technology. This is a to-do list, not a reason to feel afraid. It’s the same list this country has formed every time it discovered that those in power were lying.

Where I might be mistaken is that each case I cited is one where the truth eventually came to light. I don’t have a list of cases where the truth remained concealed because when it remains hidden, that’s how it truly “works.” Richard Helms ordered the destruction of the MKUltra files in 1973, and what we know about that program is merely what survived the shredder. Suggesting that the system corrects itself based on a sample of successful corrections is like proving that people survive plane crashes by interviewing only survivors.

Next is the issue of time. We waited forty years for the truth about Tuskegee, forty-one years for the Tonkin intercepts to be released, and over sixty years for the Guatemala experiments. “The truth comes out eventually” is a poor consolation when that eventuality arrives long after the war has ended and the victims have been buried. A self-correcting mechanism that operates on a generational lag may not be fast enough anymore. Additionally, I might be oversimplifying the consequences of targeted, personalized, and individually credible lies being accepted as truth. One falsehood broadcast to millions is something we have defenses against; however, millions of lies, each tailored for a specific individual, may pose a different challenge.

Finally, my entire argument rests on verification capacities that are currently lacking. According to the State of Local News Project, nearly 40% of all local U.S. newspapers have disappeared, leaving 50 million Americans with limited or no access to reliable local news. This trend continues to affect the media industry and audiences nationwide. Newspapers are vanishing at a rate similar to that of 2024; more than 130 papers shut down in the past year alone. Employment in journalism is steadily declining. While there has been some growth in standalone and network digital sites, these startups are largely concentrated in urban areas and have not emerged quickly enough to offset losses elsewhere. Consequently, news deserts—regions with very limited access to local news—are continuing to expand. In 2005, just over 150 U.S. counties lacked a source of local news; today, that number has exceeded 210. Meanwhile, the journalism industry is facing new and intensified challenges, including declining circulation and significant revenue losses due to changes in search practices and the rise of AI technologies. At the same time, political attacks on public broadcasters threaten to leave large portions of rural America without local news. Congressional investigative staff is a shadow of what it was in the 1970s, and FOIA requests can go unanswered for years.

However, I would rather live in a country that uncovered the truth about Tuskegee than in one that never asks questions or stops looking for the real source and evidence. This is a country that has, in fact, found out the truth, albeit late, poorly, at an enormous cost, and only because specific individuals took specific risks. This is not a reason to fear what lies ahead. It is a reason to understand who we must be.