I was all shook up when I read Hany Farid, a digital forensic and deepfake expert at Dartmouth, telling The New York Times that he doesn’t trust anything and that he “no longer trusts his own eyes.”
In a few weeks, voters across the country will head to the polls for the first midterms of the second Trump term. Before anyone decides who to vote for, there’s a more basic question that crops up. If something has happened, has it actually happened? That used to be the easy part of democracy. Today it isn’t.
The trouble didn’t start with AI, as Dr. Farid might lead you to believe. It started with a phrase. President Trump repeated the words “fake news” so often, and so effectively, that they stopped meaning, “this particular story got something wrong,” and started meaning, “don’t trust journalism, period.” That’s a shrewd move, if you think about it, because it doesn’t require proving anything false. It only requires making people doubt whoever’s telling them. And once the trust goes, evidence doesn’t mean much. An election season is exactly the time when that sentiment, the lingering doubt, gets exploited the most. It’s easy to sow doubt, and hard to clear it up.
Trump also figured out he didn’t need reporters standing between him and the public anymore. He could just post directly on Truth Social and set the day’s news before most of the country had smelled the coffee. No editor asking follow-up questions. No fact-checker holding a claim back until it checked out.
Elon Musk got to a similar place by a different road. Rather than talking around the press, he bought a piece of the infrastructure that talking runs through, Twitter, now X. His argument was that platforms should allow more speech rather than less, which is a fair point on its own terms. But then Musk built his AI chatbot, Grok, directly into X, and something genuinely new happened. Someone posts a statement on X. Grok reads it, because Grok lives right there on the same platform. A user asks Grok whether the statement is true, and the answer lands back in that same feed, seconds later. Grok isn’t just watching what happens on X — it’s part of what’s happening on X. The Musk information loop, the merry-go-round!
That’s not a hypothetical worry. In 2024, five secretaries of state, the actual officials who run our elections, wrote to Musk after Grok told users that ballot deadlines had already passed in nine states, within hours of Joe Biden dropping out of the presidential race. None of it was true. The claim circulated for over a week before X got around to fixing it. And the pattern hasn’t gone away. Reuters found in February 2026 that Grok was still producing content its own safeguards were supposed to catch.
Here’s the deeper problem, and it’s bigger than any single platform or politician. A photograph used to carry weight because someone had to physically be there, camera in hand, when something happened. AI has mostly erased that guarantee. Video can now show a person saying or doing something that never occurred, convincingly enough that even real footage starts to look deniable.
Researchers have a name for this: the “liar’s dividend.” Once everyone knows fakes are easy to produce, someone caught on an authentic recording can simply say “that’s AI.” And the rest of us are left with no quick way to sort out who’s telling the truth. Even the federal government’s own researchers admit that the tools built to detect fake images and video aren’t reliable enough to lean on when it counts. That perhaps explains why Dr. Farid finds it better to live in the Green Mountains than roam the corridors of Silicon Valley-UC Berkley.
This is precisely the job journalism has always existed to do. Not producing more content; rather, sorting through it. Who said this? How do you know? Can someone independent confirm it? These questions matter more now, not less, in the age of AI. But they take real people doing real, slow work, on a deadline that rarely favors them. That’s a tough task for local newsrooms that have spent two decades losing subscribers and ad dollars to the platforms that are making the problem worse.
The First Amendment doesn’t solve any of this for us. It was never designed to. In New York Times v. Sullivan, the Supreme Court protected even flawed reporting because, as the justices put it, public debate needs room to be “uninhibited, robust, and wide-open.” In the Pentagon Papers case a few years later, the Court refused to let the government stop a story before it ran. What the Constitution has never done, and shouldn’t start doing now, is to hand government the power to decide what’s true and what isn’t. That kind of power doesn’t stay limited once you create it. History is a witness on that point.
So the weight falls back on the rest of us, readers, voters, neighbors. Ask where a claim actually came from before you pass it along. Support the local reporters who still show up at town meetings and courthouses, because they’re the ones doing the checking nobody else will. And stay a little suspicious of anything — a video, a post, a chatbot’s confident answer — that shows up without a name attached to it.
Democracy has always run on disagreement. But it cannot run on two separate versions of what happened. That’s not just a Washington problem. It’ll show up right here, on our own ballots, in about a few weeks.
Narain Batra is a study leader at the Osher Institute at Dartmouth College and lives in the Upper Valley. In September, Batra is scheduled to teach a hybrid seminar: “Journalism: It’s Not Fake News: Restoring Faith in the First Amendment in the Age of AI” at Dartmouth. Those interested in the course should contact Lisa King at osher@dartmouth.edu.