Scientists Say AI Can Plant False Memories—and It’s Slowly Decaying Your Mind

Your memory is core to who you are. This cognitive mystery preserves your past experiences, shapes your relationships, and sculpts your sense of identity—and yet it’s also remarkably unreliable. After all, there’s a reason why lawyers consider eyewitness testimony one of the worst forms of evidence.

And in the era of artificial intelligence, things are getting even more complicated. In recent months, researchers have been investigating a phenomenon known as “cognitive surrender,” in which AI users routinely offload cognitive skills to chatbots, thereby negatively impacting their cognitive abilities. For example, in one study following 1,372 participants, users incorporated misguided AI reasoning roughly 75 percent of the time in their work. Another 2026 paper, written by scientists from Carnegie Mellon, MIT, Oxford, UCLA, and UC Berkeley, also found that people who had relied on AI performed worse at tasks than people who had never used LLMs. Using 30 months of data on more than 25,000 Chinese students, researchers reported in June 2026 that AI improved homework scores while harming secure exam results, showing that reliance on the technology broadly impacted deeper learning.

But that’s only the beginning of experts’ worries about humanity’s over-reliance on its new technological creation. A growing body of evidence also suggests that AI has the ability to tamper with our memories. Pattie Maes, a professor at MIT’s Media Lab, told The Washington Post about new findings from her yearslong investigation into how AI affects human cognition. Maes and her team used AI-generated images and videos derived from real photos. Groups of experimental subjects were then shown either the real photos, the AI-altered images, or the AI-generated videos. Those exposed to AI regularly experienced more memory distortion than those who were not. They were also more confident in those altered memories than members of the control group were.

“I think of all this as very dangerous, basically,” Maes told The Washington Post.

In another study, Maes and her team allowed three groups of participants to write an essay with no AI help, with Google search help, or with help from ChatGPT. The team measured brain activity with electrodes: solo writers showed the most activity, the search group less, and the AI group the least. After completing the essay, the AI group also struggled to remember what they wrote compared to the two other groups.

Experts have long worried that the rise of artificial intelligence and deepfakes would make misinformation even more rampant online, and research now confirms the human brain is deeply vulnerable to it. Gideon Nave, a researcher at the University of Pennsylvania, told The Washington Post that this type of research points to an AI dystopia not frequently discussed in CEOs’ long, rambling screeds about the end times, or in the action-packed plots of Hollywood sci-fi films. Rather than being overtaken by a superintelligence, humanity may simply decay under its own reliance on the technology it created.

“What we are more likely to see is humans starting to decay instead of AI taking over,” Nave said.

That doesn’t mean that AI is completely useless, of course. Its ability to help generate ideas, gather together primary sources, and refine drafts—tasks related directly to an LLM’s core abilities—have been shown to be genuinely useful. A 2024 study in the journal Nature found that when humans were better than an AI at a certain task, the combination of the two produced better results, but the opposite was true if AI could outperform the human. That’s likely because those who’d mastered their craft or skillset could more easily question an AI’s output than those who hadn’t.

The AI era may sort us into those who thrive and those who wither under “cognitive surrender” and an endless barrage of misinformation. The difference could come down to one habit: questioning your AI companions at every turn.