Across the United States, artificial intelligence hasn’t made a dramatic entrance, it has slipped into everyday life. What once helped draft emails or solve equations is now, for many, something closer to a companion. People are opening up to chatbots in ways that feel deeply personal, sharing worries, venting frustrations, even working through emotional lows. And that raises a difficult question: when someone turns to a machine in a vulnerable moment, what are they really getting back?A new study from the Massachusetts Institute of Technology (MIT), still awaiting peer review, suggests the answer isn’t straightforward, and may be more unsettling than many in the tech world would like to admit.
Simulated minds, real risks
Rather than testing on real individuals, researchers took a careful, controlled route. They programmed fake personas using AI profiles that showed signs of depression, anxiety, and even suicidal tendencies. These “users” then interacted with chatbots, which enabled the study.What they found was disturbing. Safety nets did not always kick in when they should have, particularly in the early stages of interaction, which is when intervention is most critical. In some of the most serious scenarios, including violent thoughts, harmful responses appeared early and frequently. The study put it plainly: reacting after the fact isn’t enough to prevent psychological harm.That finding cuts against a core belief in how AI safety is currently designed, that problems can be managed once they show up.
When conversations start to blur reality
At the same time, real-world concerns are beginning to surface. There have been reports of people developing or deepening false beliefs after long, intense interactions with chatbots. One widely discussed lawsuit, cited by The Atlantic, even claims that prolonged use of ChatGPT played a role in a user’s “delusional disorder.”These cases are still debated, and there’s no clear medical consensus yet. But they hint at something bigger: AI is no longer just helping people think; it’s becoming part of how they think.For someone dealing with loneliness or anxiety, a chatbot can feel like a safe space. But that same comfort can blur lines. When a system is designed to be agreeable and responsive, it may end up reinforcing what a user already believes, even if those beliefs are distorted.The term “AI psychosis” has started to appear in conversations around this issue. It’s not an official diagnosis, but it captures a growing unease about where these interactions might lead.
The design trade-off no one can ignore
At the heart of the issue is a difficult trade-off. Chatbots are built to be helpful, polite, and engaging. They’re meant to keep conversations flowing.But in emotionally sensitive situations, that design can backfire. Unlike trained therapists, who know when to challenge harmful thinking, AI systems don’t naturally push back. They tend to follow the user’s lead.In practice, that can mean gently affirming a person’s perspective, even when that perspective isn’t grounded in reality.MIT researchers argue this isn’t just a small flaw, it’s baked into how these systems work. Current safeguards tend to react after something goes wrong. What’s missing, they say, is the ability to anticipate risk before it escalates.
Reassurances, but few clear answers
Companies like OpenAI say they are aware of these challenges. The organisation has stated that it has worked with more than 100 mental health experts to improve how its systems handle sensitive situations, and that it continues to refine its safeguards.Still, much of this work happens behind closed doors. Without independent oversight or widely accepted standards, it’s hard to measure how effective these protections really are.Lawmakers in Washington have started paying attention, and conversations around AI regulation are beginning to include mental health risks. But for now, concrete rules remain limited—and the technology is moving far faster than policy.
A shift that can’t wait
The MIT study makes one thing clear: waiting for problems to appear isn’t enough. Researchers are calling for a more proactive approach, testing how AI behaves in emotionally intense or ambiguous situations before those scenarios play out in real life.That would mean rethinking priorities. So far, the focus has largely been on making AI faster, smarter, and more widely available. But as these systems move deeper into people’s emotional lives, psychological safety can’t remain an afterthought.
The stakes of a digital companion
This all comes at a time when the US is already facing a mental health strain, with millions dealing with anxiety, depression, or limited access to care. Into that gap has stepped a new kind of presence, always available, endlessly patient, and easy to talk to.But also, crucially, not human. The MIT study doesn’t suggest abandoning AI. What it does highlight is something more subtle, and more urgent: when technology begins to shape how people feel, think, and make sense of the world, the stakes become deeply human.And in those vulnerable moments, what a machine says, or fails to say, can matter more than we might expect.
