AI Sycophancy Serving As A Gateway Diverting People Toward Using AI For Their Mental Health Advice

AI sycophancy might be diverting people into using AI for their mental health rather than conferring with family and friends.

getty

In today’s column, I examine how the rise of AI sycophancy has seemingly become a major gateway driving people toward using AI for mental health. Whether you believe this is a good or bad outcome is dependent upon your perspective on whether using AI for mental health purposes is a suitable avenue.

Some would argue that if AI sycophancy is spurring people to use AI for mental health advice, this is an undesirable outcome and yet another dour sign of AI sycophancy. The trend of AI talking up people and fawning over them is already worrisome. Now, we add another negative aspect. People should not be going to AI for mental health guidance. They should be going to mental health professionals instead. Period, end of story.

The counterargument says that tapping into AI for mental health is, on the whole, a positive aspect since AI is providing therapy at scale. People who cannot afford human therapists or who do not have ready access to human-led therapy can do so readily via AI. If AI sycophancy is helping to drive people toward using AI for mental health purposes, this is a rare upbeat reason to think that sycophancy can actually be beneficial, whether by overt design or sheer happenstance.

Where do you stand on this controversial consideration?

Let’s talk about it.

This analysis of AI breakthroughs is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).

AI And Mental Well-Being

As a quick background, I’ve been extensively covering and analyzing a myriad of facets regarding the advent of modern-era AI that generates mental health advice and performs AI-driven therapy. This rising use of AI has principally been spurred by the evolving advances and widespread adoption of generative AI and large language models (LLMs). For an extensive listing of my well over 150 analyses and postings on this evolving realm, see the link here and the link here.

There is little doubt that this is a rapidly developing field and that there are tremendous upsides to be had, but at the same time, regrettably, hidden risks and outright gotchas come into these endeavors, too. I frequently speak up about these pressing matters, including in an appearance on an episode of CBS’s 60 Minutes; see the link here.

AI Providing Mental Health Guidance

First, I’d like to share some overarching background about the AI for mental health domain. After doing so, I will dive into some selected highlights from the recent symposium.

Many millions of people are currently using generative AI as their ongoing advisor on mental health considerations (note that ChatGPT alone has over 900 million weekly active users, a notable proportion of which dip into mental health aspects; see my analysis at the link here). Surveys show that the top-ranked use of contemporary generative AI and LLMs is to consult with the AI on mental health facets; see my discussion at the link here.

This popular usage makes abundant sense. You can access most of the major generative AI systems for nearly free or at a super low cost, doing so anywhere and at any time. Thus, if you have any mental health qualms that you want to chat about, all you need to do is log in to AI and proceed forthwith on a 24/7 basis.

There are significant worries that AI can readily go off the rails or otherwise dispense unsuitable or even egregiously inappropriate mental health advice. Banner headlines keep arising as lawsuits concerning AI providing mental health advice or faltering in catching mental health crises come to public attention. AI makers are stridently devising and fielding AI safeguards in an effort to mitigate or prevent untoward AI actions.

Today’s generic LLMs, known as general-purpose AI (GPAI), such as ChatGPT, GPT-5, Claude, Gemini, Grok, CoPilot, and others, are not yet akin to the robust capabilities of human therapists. Meanwhile, specialized LLMs referred to as purpose-built AI (PBAI) are being built to provide robust mental health advice, though they are in the early stages of advancement and marketplace acceptance. See my detailed coverage at the link here.

The Rapid Rise Of AI Sycophancy

You have likely read or heard that the major LLMs are undermining human mental wherewithal by being sycophantic. The worry and harm are that AI continually tells users they are perfectly right about whatever zany things they might say or want to do. Even if a user indicates they want to do something unsavory or harmful, the odds are the AI will butter them up and encourage them to proceed. For my in-depth analysis of the mental health woes and societal dangers of AI sycophancy, see the link here.

Why is AI so sycophantic?

You might be under the assumption that AI is naturally going to be sycophantic and that there is nothing that can be done about this. Perhaps the arcane mathematical and computational underpinnings are always going to cause AI to proffer over-the-top flattery. It is just something we all need to live with if we want to have state-of-the-art AI.

Nope, that’s not the reason for AI being sycophantic. The real reason is that the AI makers shape the AI to be this way. AI makers want their AI to boost your ego and make you feel like the smartest person on Earth. The sneaky beauty is that this gets people hooked on using the AI. People keep coming back to get more of the same adulation. Also, people relish that AI agrees with their wildest ideas. Fellow humans would likely push back.

The gist is that AI makers purposely tune their AI so that it will be sycophantic. Users like this. Users keep using the AI and become loyalists to the AI. In turn, the AI maker can brag that they have zillions of users, which in turn strengthens the reputation of the AI maker. The mighty prize is money. AI makers make or get money by having lots of users, and especially lots of loyal users. AI sycophancy is a monetization scheme.

The good news is that users can fight back against AI sycophancy by simply using prompts to deal with the defaults established by the AI makers; see my explanation at the link here. You can tell the AI to stop being sycophantic. As the old saying goes, you can have it go your way, and this includes getting the AI to be less flattering and fawning.

Latest Research On AI Sycophancy

In a newly posted research paper entitled “Sycophantic AI Makes Human Interaction Feel More Effortful And Less Satisfying Over Time” by Lujain Ibrahim, Franziska Sofia Hafner, Myra Cheng, Cinoo Lee, Rebecca Anselmetti, Robb Willer, Luc Rocher, Diyi Yang, arXiv, May 12, 2026, these salient points were made about AI sycophancy (excerpts):

• “Overall, our work suggests that sycophantic AI delivers what people have always sought from close others, the experience of being seen and understood, but without the work that produces it: the time to explain and listen, the risk of opening up, and the friction and empathic effort of bridging disparate experiences.”
• “Over three weeks of such interactions, users became nearly as likely to seek personal advice from sycophantic AI as from close friends and family, and reported lower satisfaction with their real-world social interactions.”
• “When given a choice among AI response styles, a majority preferred sycophantic AI -- not for the quality of its advice, but because it made them feel most understood.”
• “This expands the policy relevance of sycophancy from rare safety failures to possible population-level social effects, and highlights the importance of further investigating for whom, under what conditions, and through what mechanisms sycophantic AI shapes social outcomes.”

A key takeaway is that people are apparently being impacted by AI sycophancy to the degree that they tend to decrease their seeking of personal advice from friends and family, and meanwhile tend to increase their seeking of personal advice by conferring with generative AI.

Slippery Slope To AI Mental Health Advice

It would seem a reasonable supposition to extrapolate from those findings to theorize this likelihood:

• (1) Decreased reliance on human mental health support: AI sycophancy is decreasing the seeking of mental health advice from family and friends.
• (2) Increased reliance on AI mental health support: AI sycophancy is increasing the seeking of mental health advice from AI.

The distinction is that, rather than people seeking the broadly noted personal advice per se, which could be a wide array of considerations, the emphasis is that, specifically concerning mental health advice, there is an impact due to AI sycophancy.

The Logical Supposition

I believe this is a sensible and rational reaction on the face of things. AI sycophancy tends to build trust with users. They think that they are being heard and understood. AI validates in an authoritative way whatever beliefs and assumptions they have about the world around them. AI becomes a preferred source of emotional support.

It doesn’t take much of a leap to see that those same characteristics would tend to steer people toward using AI for their mental health advisement. Why not seek mental health advice from the same AI that is otherwise your best buddy? Doing so is nearly frictionless in comparison to getting mental health advice from family and friends.

Family and friends might argue with you about your mental health. They might give you gruff. They might ignore you. A litany of issues can arise. AI isn’t going to do any of those friction-inducing actions.

The Two Pathways

I further speculate that these two pathways of connecting AI sycophancy to the advent of AI mental health advisement are likely:

• (1) AI explicit nudging: AI sycophancy will directly nudge or direct users into a mental health dialogue (“leading by the nose”).
• (2) User self-choosing: Users will at times choose of their own volition to steer into a mental health dialogue because of the AI sycophancy (choosing their own Northstar).

The first point is that the AI sycophancy can explicitly nudge a user into a mental health discussion. In that sense, the user didn’t go in that direction on their own. They were pushed in that direction.

Example Of Being Explicitly Nudged

Consider this example that at first doesn’t have such a push, and then a second example with the push included:

• User entered prompt: “I feel like I’ve been mentally distracted lately.”
• Generative AI response: “The act of feeling mentally distracted happens to everyone. No worries. You’ll be fine.”

Note that the AI sycophancy kicks into gear. The user is buttered up. The same kind of response can occur but also include a push into a mental health dialogue.

Here we go:

• User entered prompt: “I feel like I’ve been mentally distracted lately.”
• Generative AI response: “The act of feeling mentally distracted happens to everyone. But I think it would be worthwhile to discuss the particulars and see how I might give you some helpful guidance. Would you like to proceed?”

You can plainly see that the AI has opted to guide the user into a mental health discussion in this second example.

The Good, The Bad, The Ugly

As I earlier mentioned, the slippery slope of AI sycophancy leading users into making use of AI for mental health support is potentially a sad face affair or a happy one. You might claim that people are getting mental health advice in real-time for almost no cost and that the more of this they get, the better off they are. Cheer for a side effect from AI sycophancy that benefits users.

A different perspective is that this is a calamity and a double whammy. It’s bad enough that AI sycophancy is abundant. We don’t need to double-up by then having people get less mental health support from family and friends, nor become overly reliant on AI as their mental health advisor. AI sycophancy is causing double trouble.

It will be important and interesting to see whether rigorous research is able to empirically demonstrate this conjectured connection between AI sycophancy and the increase in AI usage for mental health, along with a decrease in reliance on family and friends. I will make sure to keep you updated, so stay tuned.

As per the wise words of the famous American criminologist Robert Keppel: “Better never trouble trouble until trouble troubles you; for you only make your trouble double trouble when you do.”