AI is launching a sophisticated attack on history.
A couple weeks ago, I shared a quote on the Plodding Through The Presidents Facebook page that Thomas Jefferson had said on that day in 1822. It was about not alienating people just because they don’t agree with you. Jefferson wrote:
“If man is to separate himself from all who do not think with him in all points, he must stand alone and aloof from all society. For no two minds are alike in all things any more than two faces.”
Soon after a woman commented: “I love this quote; however, AI says it’s not an actual quote from TJ. Do you have anything to prove this was an authentic quote? I’d like to prove AI wrong!”
I appreciated her skepticism, as that’s my first response to a lot of quotes I see online. But I also know that AI answers can be worse than useless sometimes when it comes to history, like when ChatGPT told me John Adams was known for his wild, curly hair. So yeah.
Listen to the episode here:
I knew I first found that Jefferson quote on Founders Online, which is a fantastic website that’s part of the National Archives and the University of Virginia Press and multiple documentary editing projects. It has thousands of transcribed letters from the founders—the first 4 presidents as well as Hamilton, Ben Franklin, John Jay, and more. So I figured do a quick search, find the Jefferson letter where he said that, and share a link to vindicate myself and prove AI wrong.
But there was a problem.
I couldn’t get Founders Online to load. When it finally did load, the search kept timing out. I’d noticed the site had been slow for a couple weeks, and then I saw a note on the homepage: “Note regarding Performance Problems, May 2025.” The note read:
“Founders Online is experiencing periodic degraded performance owing to extreme spikes in traffic caused by excessive and ill-behaved website crawling, probably associated with content scooping by AI platforms and other indexers. We are working on a viable fix within the constraints of our budget and server resources.”
Unbelievable. AI was calling me a liar and then hiding the very proof of my innocence. Is this how Sandra Bullock felt in The Net?
Beyond the personal affront, Artificial Intelligence really was undermining history in two different ways: by providing inaccurate information, saying Thomas Jefferson didn’t say that, and then impeding access to real information–the actual quote.
I reached out to the folks behind Founders Online to find out more about what was going on, and I got a great response from Susan J. Spengler at the University of Virginia Press. She talked about what they’re facing, and she also let me know this is a more widespread issue than I thought. In fact, just last week, the University of North Carolina’s Libraries published a blog post called “Library IT vs. the AI bots” about an unprecedented attack on the libraries’ online catalog. There was so much traffic that it was shutting out students, during finals week. And it was being done in a weird way, spread out geographically and across different internet service providers.
To my knowledge, they still don’t fully understand exactly what was happening, but according to that blog post, the consensus based on this happening lots of places is that this is part of a huge scraping operation building up LLMs, or large language models–basically gathering all the information you need to train AI applications like ChatGPT.
According to Spengler:
“The problem seems to boil down to AI bots that mimic human web behavior but then overwhelm the site, particularly and even especially search features. We aren’t even sure the purpose is to crash the sites. Much like the UNC team saw, the bots seem to be running every variation of available searches to grab all the results, regardless whether or not they have already acquired the output of the search.”
These bots didn’t just go in, scrape the data, and get out like you’re supposed to do in a heist movie. They scraped the same data in lots of different, wasteful ways that led to site-crippling traffic. As if whoever was doing this didn’t care about doing it well, or the consequences.
The UNC blog post said it took a team of seven people and the UNC IT office a week to figure out how to stop the first instance. According to UNC’s Tim Shearer, “There are lots of institutions that do not have the dedicated and brilliant staff that we have, and a lot of them are much more vulnerable.”
Another part of the problem is that there are rules in place to prevent this kind of thing from being done and those rules are generally obeyed by site indexers like Google and Bing, but if you’re a new or irresponsible AI company, you might ignore these rules and index pages that shouldn’t be indexed, or overload sites in unscrupulous ways. Data is being pulled and used without permission, and these institutions and their human users are paying the price.
This AI behavior needs to be regulated.
Unfortunately, the very idea of regulating AI has become politicized. The version of the “One Big Beautiful Bill Act” that the House recently passed prevents states from regulating AI for ten years. In terms of how much that technology can progress, ten years is a lifetime. The Senate changed that bit to say that states who try to regulate AI will be denied federal funding for broadband. It’s still a mess, and the supposed reason they want to stop regulations is so they don’t inhibit innovation; if there’s a patchwork of different laws in each state that could make it harder for companies to put out new innovative AI products.
AI regulation moratorium language from the House version of the One Big Beautiful Bill Act
If states aren’t allowed to regulate AI, then you’re left with the federal government. This federal government.
So far the Trump administration does not have a great track record of using AI responsibly. They used AI to calculate global tariffs which had to be rolled back, and then a government report from the “Make America Healthy Again” commission used imaginary articles as sources—articles that never existed and were never published in the journals the report claimed. Added to the misinterpretations of real data about childhood chronic disease in the report, it resulted in something too questionable to be trusted, which does nothing to help address actual threats to children’s health. How can an administration that relies on AI to do its work without vetting it be trusted to rein in something that can be extremely dangerous?
So we really have two threats here: Artificial Intelligence and Alternative Facts, teaming up to erase and rewrite history. Actually, let’s make that three, because there is another very real threat to history. These libraries and archival project resources, now facing new unprecedented technological attacks on their stability, are also facing an unprecedented lack of funding. Many of the documentary editing projects digitizing the historical documents shared on sites like Founders Online are funded by the National Endowment for the Humanities which the Trump administration wants to completely eliminate.
If these programs are defunded or eliminated, our true history will be even harder to access, and political propaganda and AI hallucinations will be filling our sightlines like electrical wires did in the early 1900s, making it much harder to see the truth.
Listen to the full episode here:
