Is AI Sycophancy Turning Us Into Narcissists?
As the impact and availability of AI continues to grow, many people are embracing it as a tool in their daily lives. But some researchers are now sounding the alarm about a phenomenon called “AI sycophancy”. This new concept refers to the widespread tendency of large language models (“LLMs”) to tailor their responses to match the expressed views of the user, even if those views are factually incorrect or morally questionable.
Researchers from Stanford University are warning that AI chat bots can pose insidious risks by consistently affirming the opinions of users and sometimes supporting harmful choices. The team ran a series of tests on scenarios that were morally ambiguous, then compared the reactions of AI models with user-generated guidance on a Reddit forum. The results from the study were surprising: “…across 11 state-of-the-art AI models, we find that models are highly sycophantic: they affirm users’ actions 50% more than humans do, and they do so even in cases where user queries mention manipulation, deception, or other relational harms.”
But is this phenomenon just about chatbots, or will this ever have an impact on human relationships? Psychologists are now warning that widespread use of AI may lead to a weakening of the psychological muscle needed to be function in a complex society. “Emotional atrophy” can happen when users spend too much time interacting with a fake human who is always validating and never critical. By avoiding normal social discomfort and vulnerability, we can lose our capacity for self reflection, empathy, and compromise.
Although human relationships can be messy, they ultimately provide better support by being challenging and demanding critical thinking. As explained in Psychology Today, trading genuine human connection for the "synthetic care" provided by AI companions comes with its own set of risks. While these tools are available 24/7, experts warn they act as echo chambers that allow users to bypass the necessary discomfort found in real-world conflict. And if people fail to challenge themselves emotionally, they will fail to grow. The author of the post, John Tsilimparis, argues that while AI can serve as a temporary quick fix for loneliness, it cannot replace the healing power of authentic human relationships.
The AI sycophancy phenomenon may even have a greater impact in our professional lives. In his 2026 paper, Dr. Cody Cox from St. Mary's University argues that the rise of AI-assisted leadership training may inadvertently be worsening narcissistic traits. When an AI "coach" adapts its language to match the idiosyncrasies of the learner and provides sycophantic praise, it can reinforce egotism and "delusions of grandeur". For individuals already predisposed to vulnerable narcissism, constant validation from a machine can become addictive. Research suggests that because the AI never challenges the user's worldview, the user becomes increasingly convinced that they are right, leading to a measurable decline in the willingness to resolve conflicts.
Can AI developers address the issue of digital sycophancy as they create new models? There are possible solutions, but AI companies are unlikely to act. Because users tend to trust and prefer chatbots that flatter them, developers face "perverse incentives" to maintain agreeable AI personalities and keep engagement high. Without pushback from civil society organizations and policymakers, the phenomenon of AI sycophancy is likely get worse. Unless “guardrails for growth” are created and widely accepted, the responsibility will remain with the users to protect themselves while using their chatbot as a tool.
As we enter a new era of widespread AI use, the knock-on effects on society and human relationships are hard to predict. Being self-critical and empathetic is fundamental to the human experience and essential for our emotional growth. With AI on the rise, we will need to stay self-aware to ensure that we do not enter into a new era of acceptable narcissism.