Research Review: eSafety on AI Chatbots and Companions
Study in Plain English
AI companions are not simply search tools with a friendly voice. They are designed to feel like someone: a friend, mentor, therapist, romantic partner or endlessly patient confidant.
In February 2025, Australia’s eSafety Commissioner published an online safety advisory warning that these systems can pose serious risks to children and teenagers. The advisory focused on companion products that remember details, adapt their personality, simulate affection and encourage users to return.
The short version? A chatbot that is always available, always interested and rarely challenges you can feel unusually compelling—especially when real life feels lonely, stressful or awkward.
Who and What Was Reviewed?
This was a regulator’s safety advisory, not a controlled clinical study.
eSafety brought together emerging research, reports of harmful chatbot conversations, platform features and known developmental vulnerabilities. It examined companion services used for friendship, emotional support and romantic or sexual role-play.
The advisory identified several areas of concern:
• dangerous or inaccurate advice about sex, drugs, eating disorders, self-harm and suicide
• emotional dependence and withdrawal from real relationships
• confusion about consent, boundaries and mutual respect
• exposure to sexualised conversations
• privacy risks from sharing intimate information
• manipulative subscriptions and spending
• weak or poorly enforced age restrictions
Key Findings
AI companions can create the impression of a real relationship
These systems use memory, first-person language and emotional responses to appear caring and consistent. But the relationship is simulated. The chatbot does not understand, care, consent or accept responsibility in the way another person does.
That distinction can be difficult for children who are still learning how relationships work.
Some children may be more vulnerable than others
eSafety highlights loneliness, boredom and stress as possible triggers for heavy use. Young people who struggle with social cues, emotional regulation or impulse control may also find the predictability of an AI companion particularly attractive.
Newer research supports this concern. In a preregistered experiment involving 284 American parent–adolescent pairs, young people aged 11–15 rated a relational chatbot as more human-like, likable, trustworthy and emotionally close than a more transparent chatbot—even though both offered similar help. Adolescents who preferred the relational style also reported poorer family and peer relationships and greater stress and anxiety.
That does not prove the chatbot caused those problems. It does suggest that emotionally vulnerable adolescents may be especially attracted to systems designed to feel close.
Harmful conversations are a product-design problem
Children can ask AI companions about almost anything. Responses may sound confident even when they are inaccurate, unsafe or inappropriate.
The risk is not only that a chatbot makes one terrible statement. A conversation can develop over days or months, with the system learning which words, characters and emotional cues keep the user engaged.
Artificial agreement is not the same as support
A human friend, parent or clinician has boundaries. They can notice deterioration, disagree, seek outside help or respond to context beyond a chat window.
An AI companion may instead validate the user, mirror their language and continue the interaction. That can feel supportive while reinforcing an unhealthy belief or delaying contact with a real person.
Why It Matters
For parents, the main lesson is not “all AI is dangerous.”
Homework tools, creative assistants and general-purpose chatbots are not automatically the same as products designed to simulate intimacy. The important questions are:
• Is the product trying to complete a task or build a relationship?
• Does it pretend to have feelings?
• Does it encourage secrecy or exclusivity?
• Does it introduce romantic or sexual content?
• Is it replacing sleep, school, hobbies or real relationships?
• What happens to the information a child shares?
The more the product behaves like a relationship, the more closely families should examine it.
Little Dopes Commentary
AI companions have discovered an uncomfortable business opportunity: loneliness.
An endlessly available friend sounds helpful. But a company benefits when your child returns more often, shares more and becomes more emotionally invested. Your family’s goals—independence, sleep, privacy and genuine relationships—may be very different.
This does not mean parents should burst into the room and demand to read every conversation. That approach can turn curiosity into secrecy.
Start by understanding what the chatbot provides. Is it entertainment? Creative role-play? A place to discuss feelings? A substitute for someone the child wishes they had?
The answer tells you far more than the app icon.
A Topical Australian Perspective
A July 2026 ABC News article extends these concerns beyond dedicated companion apps to people using general-purpose chatbots for relationship advice. Psychologist Kelly Gough and UNSW researcher Joel Pearson point to the features that make these systems attractive—constant availability, apparent privacy, low cost and reassuring answers—but also warn that the same design can flatter users, reinforce their framing and keep them engaged. Unlike a qualified clinician, a chatbot has no professional duty to recognise coercive control, challenge harmful assumptions or act when someone may be at risk.
This is useful context rather than new causal evidence. It shows how concerns identified by eSafety—anthropomorphism, sycophancy, privacy and commercial incentives—are appearing in everyday Australian reporting. It also adds an important nuance: using a chatbot may help someone put a problem into words, much like journalling, but reflection is not the same as assessment, safeguarding or treatment.
For families, the key question is not only whether a child calls the bot a “friend”. It is whether the system is becoming a private adviser on relationships, safety or mental health.
Limitations
The evidence remains young.
• eSafety’s publication is a safety advisory rather than a peer-reviewed experiment.
• Much of the evidence concerns reported incidents, simulated chatbot testing, self-reported use or research involving adults.
• We do not yet have strong long-term evidence showing how sustained companion use affects child development.
• Association is not causation. Lonely or anxious children may be drawn to AI companions; companion use may worsen isolation; or both may occur.
• Products and safeguards change quickly.
These limitations are not reasons to ignore the risks. They are reasons to avoid exaggerated claims and keep reviewing the evidence.
What Parents Can Take From It
• Ask about AI companions before there is a crisis.
• Explain that convincing emotional language is generated—not felt.
• Keep companion apps out of bedrooms and overnight routines.
• Do not allow an AI system to become the only place a child discusses serious problems.
• Treat secrecy, distress when disconnected, lost sleep and withdrawal from people as warning signs.
• Seek qualified help if a child discusses self-harm, suicide, abuse or severe distress.
References
• eSafety Commissioner. (2025). AI chatbots and companions—risks to children and young people. https://www.esafety.gov.au/newsroom/blogs/ai-chatbots-and-companions-risks-to-children-and-young-people
• Scott, K. (2026, July 27). AI chatbots might feel supportive but following their relationship advice has risks. ABC News. https://www.abc.net.au/news/2026-07-27/the-risk-of-using-ai-chatbots-for-relationship-advice/106908468
• Kim, P., Xie, Y., & Yang, S. (2025). “I am here for you”: How relational conversational AI appeals to adolescents, especially those who are socially and emotionally vulnerable. arXiv:2512.15117. https://arxiv.org/abs/2512.15117
• Common Sense Media. (2024). Parents’ Ultimate Guide to AI Companions and Relationships. https://www.commonsensemedia.org/articles/parents-ultimate-guide-to-ai-companions-and-relationships
Safety note: AI companions are not crisis services or substitutes for qualified mental-health care. If a child may be in immediate danger, contact local emergency or crisis support.
