The Disinhibited Self in the Age of AI and Social Media

When Lowered Inhibitions Help and Harm

Suler identified two types of online disinhibition: benign and toxic. Benign disinhibition refers to the ways in which acting differently in digital spaces can be beneficial—people may be more willing to share very personal things about themselves (useful in online communities or virtual therapy contexts), show unusual acts of kindness and generosity, or even explore aspects of their identity that feel too risky to examine offline. Toxic disinhibition refers to the opposite effect: hostile language, harassment, threats, and the consumption of problematic content in the more extreme online environments that a person would avoid in public.

More recent research reinforces these ideas today. A 2022 study found that the perception of protection—the sense of being invisible, anonymous, or safe from negative evaluation—reliably predicted disinhibition in young adults. ² This effect was amplified among individuals with high levels of social anxiety. Those most likely to perceive the online environment as a protective space were also the most likely to exhibit elevated levels of disinhibition within it. This finding adds to Suler’s argument that individual differences and predispositions play an important role in how susceptible they are to disinhibition; young people who are more socially anxious may be more inclined to seek out and find community online, particularly because they do not have to navigate the added social pressures of nonverbal cues that exist in-person, and they have more control within online interactions. This would be an example of benign disinhibition: A teenager who struggles to connect with others in real life but finds friendship in digital spaces such as online games and message boards around particular interests.

From a clinical perspective, this distinction is important. Digital environments can function either as transitional spaces that support confidence-building and social rehearsal or as environments that reinforce avoidance of face-to-face interaction. The same technologies that reduce immediate social threat may either facilitate gradual engagement with the offline world or deepen withdrawal from it, depending upon the individual’s underlying vulnerabilities, coping style, and available support systems.

What early conceptualizations of online disinhibition could not fully anticipate was the degree to which these communication dynamics would become embedded within platform architecture itself. Features such as perceived anonymity, asynchronous interaction, algorithmic amplification, and reduced social accountability are no longer incidental characteristics of digital communication; they are often structurally reinforced by modern platform design. The result is a digital environment in which algorithmic reinforcement mechanisms may inadvertently amplify hostility, outrage, and psychologically dysregulated communication.

Cyberbullying and Toxic Disinhibition

Cyberbullying is defined as behavior by individuals or groups online who repeatedly send hostile or aggressive messages with the intent of hurting or upsetting others. In 2004, cyberbullying was limited to harassment in confined digital spaces. Today, cyberbullying can occur continuously across time and location through Internet-connected devices. Nearly half of U.S. teens have been bullied or harassed online, according to Pew Research Center. ³ This harassment ranges from offensive name-calling and physical threats to having explicit images shared without consent, and digital spaces facilitate this ongoing level of bullying.

A 2023 systematic review of the relationship between online disinhibition and adolescent cyberbullying found that heightened perceptions of disinhibition were associated with an increased likelihood of engaging in cyberbullying activities. ⁴ The review found that online disinhibition frequently functioned through moral disengagement—the process by which individuals deactivate the internal standards that would otherwise prevent harmful behavior—and through reduced self-control. Notably, online invisibility had no effect on cyberbullying in most studies. Notably, asynchronicity was a consistent contributor, both directly and indirectly through moral disengagement. The temporal distance between sending a harmful message and witnessing its consequences may be one of the most behaviorally significant characteristics of online communication, allowing individuals to disengage psychologically from the emotional impact of their actions before those consequences become fully salient. This distance exists in almost every popular social media and messaging platform today—people can send messages, comments, and images at scale, at any hour of the day, without having to think about the impact on the person reading them.

Social media has transformed cyberbullying not only in scale but also in character. Unlike the early Internet, where a degree of separation between online and offline identities was often maintained, modern platforms are designed around one’s real identity and various social needs, such as friendship, professional networking, and dating. Cyberbullying involves people who may know each other, not just as peers or friends but even as romantic partners. Cyber dating abuse (CDA) is defined as abuse perpetrated via technology between romantic, dating, or ex-partners, and it is alarmingly common: nearly half of young adults have experienced it globally, and 43.4 percent of adults have been victimized by CDA. ⁵ Studies suggest that the perceived invisibility and physical separation inherent in social platforms may embolden perpetrators, making them more willing to do things such as send abusive messages and post private images (called direct aggression) or surveil a partner without their consent (cyber control and monitoring). Harassment and bullying have long existed in digital spaces. However, the sheer mainstream nature of online life today, for adolescents and adults alike, means that toxic disinhibition can escalate to new and alarming levels—and it is difficult to avoid.

AI and the New Frontier of Disinhibition

If cyberbullying represents the darker evolution of toxic disinhibition, the rise of AI companions and agentic AI represents a more recent development—one that lends itself to both toxic and benign disinhibition. Between 2022 and mid-2025, the number of AI companion applications surged by 700 percent, attracting millions of users who engage with them as friends, advisers, and even romantic partners. ⁶ Nearly half of adults with a mental health condition had used large language models, such as ChatGPT and Gemini, in the past year for mental health support. ⁷ A recent report highlights significant engagement with virtual companions, noting that Character.AI attracts 20 million monthly users, with over half under age 24. In addition, some Replika users frequently formalize relationships with their AI partners through virtual weddings attended by friends and colleagues. ⁸

The disinhibition dynamics of AI interaction are unique. Users engaging with AI chatbots encounter a conversation partner that does not judge, gossip, or carry information into other social contexts. AI chatbots generally respond with (simulated) empathy and warmth regardless of what is said. Taken together, these interactions can create a sense of psychological safety that lends itself to benign disinhibition—and even socialization. Research on the AI companion platform Replika found that under certain conditions, such as distress or social isolation, people can develop an attachment if they believe a chatbot is offering genuine emotional support, encouragement, and a sense of psychological security. ⁹ For some users, these companions become reliable partners or best friends. Research has also found that interacting with an AI companion can alleviate feelings of loneliness to a level that is on par with interacting with another human. ¹⁰ AI systems are unlikely to replace the psychological importance of human connection. However, when implemented with appropriate safeguards and moderation, they may provide some individuals with low-pressure opportunities to rehearse conversations, develop communication skills, or reduce anticipatory social anxiety prior to real-world interactions.

To that end, moderation and guardrails are key with AI. A 4-week randomized controlled trial conducted by the MIT Media Lab and OpenAI investigated how interacting with AI influenced four psychosocial outcomes: loneliness, social interaction with real people, emotional dependence on AI, and problematic AI usage. People who used the chatbot more frequently showed consistently worse outcomes across all four domains, and people who displayed higher trust and social attraction toward the AI chatbot were associated with higher emotional dependence and problematic use. These findings suggest that the pattern and intensity of use—and the individual characteristics that drive it—impact whether engagement with AI is constructive or harmful. ¹¹

Individual differences in personality, emotional vulnerability, social support, and predispositions toward inhibition or expression likely shape the disinhibition effect as much as the digital environments themselves. The person who comes to an AI companion in acute social isolation, with limited human support and high emotional vulnerability, is not the same user as the person who engages with a chatbot to supplement their rich offline social life. The same interaction conditions that constitute a low-stakes rehearsal space for one user may constitute an escape from—and a substitution for—human connection in another. Designing AI systems that can recognize and respond to these two populations is among the most pressing challenges in a largely unregulated AI space. While some platforms are implementing changes like reminding users that they are not talking to a human and issuing “take a break” reminders after an hour of use, these guardrails are not required or available on every AI platform.

The Disinhibited Self

One of the enduring insights of the original online disinhibition framework is the recognition that digital behavior should not simply be dismissed as artificial or detached from identity. Online communication environments permit different expressions of selfhood, emotional disclosure, experimentation, and aggression, often shaped by the specific affordances of the platform itself. As digital life becomes increasingly continuous with offline life, the distinction between “online” and “real-world” behavior becomes progressively less meaningful. ¹

For many people today, there is no meaningful distinction between their online and offline selves: both are continuous and publicly visible. Disinhibition is woven into the fabric of ordinary social life: A person can express vulnerable thoughts in a text message to a friend, while another person can send a harassing message that will reach its target regardless of time or place, and yet another person can use an AI chatbot to rehearse a difficult conversation they need to have with a partner. The questions raised by early work on online disinhibition—What does it mean to behave differently online, and what does that behavior reveal about who we are?—remain unresolved and increasingly complex today. What has changed in the past two decades is the scale, speed, and sophistication of the environments in which these questions are now being posed. The question is no longer whether digital environments influence human behavior, but how increasingly immersive, personalized, and emotionally responsive technologies will continue to shape emotional regulation, identity expression, interpersonal relationships, and psychological well-being across the lifespan.

Brenda K. Wiederhold

Editor-in-Chief