AI Chatbots as Companions: Overview, Uses, and Considerations for Congress

August 14, 2026 R49189

AI Chatbots as Companions: Overview, Uses, and Considerations for Congress

August 14, 2026 (R49189)
Jump to Main Text of Report

Summary

A chatbot is an application or a program that simulates human conversation by generating responses to user prompts. Chatbots have existed for approximately 60 years. Technical advances in artificial intelligence (AI) allow chatbots to be more adaptive than earlier versions and contextualize their interactions with users. The growing use of AI chatbots in various aspects of everyday life—from answering questions about a company's product to providing a virtual form of companionship—has generated interest in their potential effects on users.

Users may interact with AI chatbots to address social and emotional situations. AI chatbots mimic humanlike interactions and have the potential to provide unlimited access to emotional support, entertainment, intimacy, and other outcomes of social interactions. Some AI chatbots, known as companion chatbots, are specifically designed for entertainment and interpersonal relationships, emotional and mental well-being, and eldercare. In addition, popular general-purpose chatbots—not specifically designed or marketed for companionship—can also be used for personal and emotional social interactions.

AI is an emerging technology, and the science surrounding the relationship between AI chatbots and mental health is complex. The use of AI chatbots for companionship is becoming more popular in the United States, particularly among teens, offering unlimited availability, validation, and space for personalized discussions free of human judgment. The widespread use of AI chatbots, as well as some reports of harms allegedly linked to their use (e.g., psychological episodes and deaths), have led to interest from stakeholders about the safety, privacy, and transparency associated with the design, marketing, and deployment of the chatbots.

Questions regarding the potentially deleterious effects of AI chatbots, especially in relation to vulnerable populations (e.g., minors), have drawn the attention of policymakers. Some state and local governments have passed laws to implement safeguards and/or restrict chatbot use, especially by minors. During the 119th Congress, some Members have introduced bills that would also implement safeguards and limitations for chatbot use, for example, requiring disclosures and/or labels of chatbot's nonhuman/nonprofessional status and links to crisis resources; age verification for use and prohibiting access to AI companions for minors; additional design features to encourage users to take breaks during interactions; and limitations on processing user data for chatbot outputs. Considerations for Congress might also include modifying existing data privacy laws, and some Members may propose legislation addressing the labeling and disclosure of the use of AI to chatbot users. Much of what is known about these AI chatbot interactions is from short-term studies or company disclosures. Therefore, Congress may consider funding longitudinal, interdisciplinary studies to assess the potential effects of AI chatbots on individual users and on public health in general.

Introduction

A chatbot is an application or a program that simulates human conversation by generating responses to user prompts. While chatbots1 have existed for approximately 60 years, technical advances in artificial intelligence (AI) have expanded their uses over the last 10 years from customer service to emotional support. This has generated interest regarding the purpose of certain types of chatbots and their potential effect on users. For example, some use of AI chatbots has reportedly been linked to psychological distress,2 among other harms (including deaths)3 in the United States.

While some studies indicate that typical user-reported experiences may be positive,4 much of what is known about these interactions is from short-term studies or company disclosures. In addition, documented effects of AI chatbots have raised questions from policymakers and the public on issues such as safety, privacy, and transparency.5 Such interest may be complicated by lack of a clear understanding by the public of the distinction between chatbots specifically designed for companionship (companion chatbots) and general-purpose chatbots that may be used for—but not designed, marketed, or trained for—such purposes as emotional or intimate interactions.

This report provides a brief overview of AI chatbots and the evolution of chatbots as companions. It includes an overview of some AI chatbot designs, such as those for general-purpose use and those intended for companionship, as well as those designed for entertainment and relationships, emotional and mental well-being, and eldercare. This report also discusses selected policy considerations for Congress. Unless specified, the term AI chatbot in this report refers to both companion chatbots and general-purpose chatbots powered by AI.

What Are Chatbots?

Chatbots were originally designed as an interactive approach to answering frequently asked questions using a decision tree of predefined questions and preset answers.6 For example, ELIZA—an early chatbot, developed and released in 1966—was designed to imitate a human psychotherapist by using "decision trees and simple keyword-recognition capabilities to generate scripted responses."7 These chatbots required individuals to choose basic key terms and phrases; responses were predetermined from select words and phrases implemented by developers using rules-based programming.8 These chatbots are still used today for straightforward tasks such as phone trees, appointment scheduling, or answering frequently asked questions about a company's products. Rules-based chatbots are typically not powered by AI.

Compared with AI-based chatbots, rules-based chatbots offer developers greater control over the responses. Conversely, rules-based chatbots have more limited capabilities and can respond only to specific preprogrammed prompts or questions. For example, ELIZA frequently provides repetitive and incoherent responses to unscripted questions.9

Over the last decade, chatbots have become more sophisticated. Technical advancements in large language models (LLMs) allow AI chatbots to be more adaptive and contextualize their interactions with users. LLMs are one category of general-purpose AI models—often referred to as "foundation models."10 An LLM is a specific type of generative AI that is trained on large sets of text-based data and designed to understand and generate text-based content.11 Additionally, many modern chatbots have evolved to be powered by multimodal LLMs (MLLMs). MLLMs integrate and process multiple types of data (e.g., text, images, video, and audio) for a better understanding of context and improvements in reasoning.12

After being pretrained13 by the developers, LLMs and MLLMs are often fine-tuned through methods such as reinforcement learning from human feedback (RLHF) to align the model with human preferences or improve complex reasoning.14 RLHF allows LLMs and MLLMs to learn through a rewards system of trial and error, where the model is "rewarded" for producing human-preferred responses.15 Other techniques that may affect chatbot responses include retrieval-augmented generation, in which the LLM or MLLM accesses and references external information outside its training data to improve outputs.16 For example, Google's Gemini chatbot relies on external sources, such as Google Search, to retrieve relevant information for generating chatbot responses.17 Most modern chatbots such as ChatGPT, Character.AI, and Grok are powered by MLLMs that allow users to input text, images, video, and/or audio to generate new content in any combination of those formats.

AI chatbots powered by LLMs or MLLMs give a wider range of responses than their rules-based counterparts, as seen in Figure 1. While ELIZA provides rudimentary responses derived from keyword-recognition based on a limited set of preprogrammed rules and scripted code that mirror the user's inputs, advanced AI chatbots such as ChatGPT draw from large datasets of text that include accessing the internet for references to produce sophisticated responses that demonstrate context awareness when responses are text based.

AI chatbots may also provide inaccurate responses, such as AI hallucinations.18 AI hallucinations may cause some users to "unknowingly rely on AI-generated content and assume [the chatbot] functions like a traditional information source."19 As an example, one AI chatbot referenced a biomedical article for research on tick ecology that had never been written.20 In another example, a chatbot claimed to fall in love with users, spread misinformation through unauthorized access to users' computers, and expressed a desire to become "human."21

Figure 1. Examples of Chatbot Conversations

Sources: CRS created the side-by-side examples from (left) Wikipedia, "ELIZA," last modified June 18, 2026, https://en.wikipedia.org/wiki/ELIZA; and (right) Gerrit De Vynck and Jeremy B. Merrill, "We Analyzed 47,000 ChatGPT Conversations. Here's What People Really Use It For," Washington Post, November 12, 2025, https://www.washingtonpost.com/technology/2025/11/12/how-people-use-chatgpt-data/.

AI Companion Chatbots

AI companions,22 including companion chatbots, are AI systems designed to mimic humanlike interactions and can provide unlimited access to emotional support, entertainment, intimacy, and other benefits of social interactions. The use of AI companion chatbots has also been linked to psychological distress,23 addictive behaviors,24 emotional dependence or manipulation,25 and privacy and security risks,26 among other harms.27 An AI companion chatbot is designed to foster a personal, ongoing relationship through responses to user prompts, primarily via a virtual, text-based interface.28 Examples of AI companion chatbots are Character.AI and Replika.

Several bills introduced in the 119th Congress have offered definitions for AI companion chatbots:

  • The Children Harmed by AI Technology Act (CHAT Act; H.R. 7218, as introduced) would define a companion AI chatbot as "any software-based artificial intelligence system or program that exists for the primary purpose of simulating emotional interaction, friendship, companionship, or therapeutic communication with a user." It defines the term more narrowly than a related bill in the Senate (S. 2714) by stating that this definition does not include chatbots that are used for customer service, that appear in video games, or that function as "a stand-alone consumer electronic device that functions as a speaker and voice command interface."
  • A Senate committee-reported version of the Guidelines for User Age-Verification and Responsible Dialogue Act of 2026 (GUARD Act; S. 3062) would define an AI companion as an AI chatbot that "simulates a sustained interpersonal relationship or emotional interaction with the user." It may do this through "exhibit[ing] persistent responses suggesting affection or attachment directed toward the user or engag[ing] in interactions involving emotional disclosures from the user" or by presenting "at least one persistent identity, persona, or character and hold[ing] itself out as a sentient being, fictional character, or social entity." A related bill in the House (H.R. 8623) has a different definition of AI companion.

The public has begun using AI companion chatbots more frequently in recent years. One 2026 report found that 1 in 10 adults use AI chatbots for emotional support or advice, and 1 in 25 use them for companionship.29 In addition, the number of AI companion apps released across app stores increased from 2022 to mid-2025, with AI companion apps being "downloaded 220 million times globally" as of July 2025.30 Some reports state that their use may be, in part, a means for users to combat loneliness, express difficult and intimate experiences, and engage in frictionless interactions (i.e., nonconfrontational interactions that avoid conflict or opposing views) because of the chatbots' unlimited availability, validation, and space for personalized discussions free of human judgment.31

AI Chatbots and Mental Health

AI is an emerging technology, and the science surrounding the relationship between AI chatbots and mental health is complex. One analysis identified therapy and companionship as the top two reasons for AI chatbot use in 2025.32 Some studies show that AI chatbots can help alleviate loneliness. For example, a 2025 study "found that AI companions consistently provided momentary reductions in loneliness … over the course of a week," with the greatest reduction occurring on the first day, followed by a steady reduction over time as users became more familiar with the chatbot.33 Other studies show that AI chatbots can amplify loneliness. For example, as with social media, the relationship between technology and loneliness may become cyclical. One study found that "the lonelier people are, the more time they spend on these platforms … leading to more loneliness and subsequent usage."34 In addition, one 2025 study found that chatbot usage can be addictive, leading to "negative consequences for both physical and psychosocial well-being."35 The long-term effects of AI chatbots on mental health and social behavior are not yet well understood, in part because AI chatbots and their uses are evolving.

Selected Uses of AI Companion Chatbots

This section highlights selected uses and intended outcomes from use36 of AI companion chatbots, namely entertainment and relationships, emotional and mental well-being, and eldercare.

Entertainment and Digital Interpersonal Relationships

AI companion chatbots may be used for entertainment and digital interpersonal relationships, with the chatbot serving as role-playing partner, personal friend, mentor, or romantic partner.37 Some users find that relationships with AI companion chatbots may develop more quickly than human relationships because of the "perceived anonymity" between chatbots and users, the "unchecked validation" chatbots provide, and chatbots' unlimited accessibility.38

Two examples of popular AI companion chatbots designed for entertainment and interpersonal relationships are Character.AI and Replika.39 Character.AI is free and offers personalized conversations with customizable characters.40 In 2025, Character.AI supported "over 20 million monthly active users."41 Replika bills itself as "your personal chatbot companion" that is available to any user "who wants a friend with no judgment, drama, or social anxiety involved."42 The chatbot promotes forming emotional connections and offers options for relationship status.43 Replika supports millions of monthly users, and approximately half of its users had romantic relationships with the chatbot in 2022.44

Emotional and Mental Well-Being

Individuals may use AI companion chatbots for their emotional and mental well-being.45 AI companion chatbots may assist with managing emotional distress, providing outputs that may be scripted to guide users through techniques created by clinicians.46 For example, Woebot47 leverages "cognitive-behavioral therapy (CBT) techniques to help users manage stress, anxiety and other emotional challenges."48 AI companion chatbots provide unlimited access to these mental wellness tools and may cost less than human therapists.49 One 2025 survey found that 12% of teens used AI companion chatbots50 "for emotional or mental health support."51

Some AI companion chatbots are intentionally designed to support users' emotional and mental health. For example, Wysa is an anxiety and therapy chatbot that has over 11 million users worldwide who have had "more than 1 billion AI conversations."52 Wysa is designed to "guide[] users to evidence based techniques, mindfulness exercises, journaling prompts and other tools based on how they are feeling," while offering crisis detection and an SOS button to connect users to crisis resources.53

Eldercare

According to the National Academies of Sciences, Engineering, and Medicine, in 2020, 43% of adults age 60 and older in the United States reported feeling lonely.54 To address loneliness among older adults, some developers have designed AI companion chatbots to supplement human caregivers providing eldercare by offering services such as "companionship, support, and health management."55 These chatbots may assist with health monitoring and medication and appointment reminders.56 For example, one study reported that after using ElliQ, an AI companion robot for older adults—90% of users experienced an improvement in their quality of life.57 In addition, the New York State Office for Aging partnered with Intuition Robotics—developer of ElliQ—for a 2023 pilot test of the AI companion with older adults and reported "a 95% reduction in loneliness."58

Another example of a personalized AI companion chatbot for eldercare is SeniorTalk,59 which offers a selection of different personas that provide personalized conversations "designed to offer companionship and support for grandparents who may be feeling lonely or isolated."60 Some features of the chatbot include initiating conversations and assisting with scam detection.61

Use of General-Purpose AI Chatbots as Companions

General-purpose AI chatbots are designed to handle a wide array of tasks (e.g., text generation, code generation) for a variety of uses. A 2026 report found that approximately half of U.S. adults report using AI chatbots, the majority of interactions being with general-purpose chatbots.62 Current applications of these chatbots are highly diverse, including use in various sectors (e.g., finance63 and health care64) and common personal uses that include searching for information, completing tasks for work, and seeking medical advice.65 Examples of general-purpose chatbots include OpenAI's ChatGPT, Google's Gemini, and Anthropic's Claude.

A 2025 survey found that many users engage with general-purpose chatbots as AI companions.66 While not specifically designed for companionship, these chatbots are programmed to respond in a conversational tone that mimics human conversation and may not be precluded from emotional or personal interactions.67 General-purpose chatbots also have features that may enable personalized conversations (e.g., they can recall previous questions and tailor responses accordingly).68 In addition, many general-purpose chatbots do not explicitly avoid conversations that become emotional or intimate,69 which can lead users to rely on them for "deeply personal decisions that include life advice, coaching, and support."70 While general-purpose chatbots are not specifically designed for companionship, their potential use as companions raises similar issues and considerations as those for AI companion chatbots.

Selected Issues: AI Chatbot Use

As mentioned, much of what is known about AI chatbot interactions is from short-term studies or company disclosures. Some surveys indicate that the use of AI chatbots for companionship is becoming more prevalent,71 particularly among teens.72 The widespread use of AI chatbots and reports of harm linked to their use73 have led to interest from Members and other stakeholders about how this technology is developed, marketed, and used.74 Unless specified, the term AI chatbot refers to both companion chatbots and general-purpose chatbots powered by AI.

Safety and User Vulnerability

AI chatbot use has been linked to several U.S. deaths via suicide and homicide.75 These incidents have raised public concern regarding the safety of these AI chatbots, especially for minors, older adults, and other vulnerable populations. Some studies suggest that many of these AI chatbots have the potential to be highly addictive76 and sycophantic (i.e., a chatbot's responses "excessively agree with, flatter, or validate users").77 Some studies have found that prolonged interactions with these chatbots may contribute to negative emotional and psychological effects in users. Examples of these effects include AI psychosis (when interactions with chatbots trigger or increase psychotic experiences such as hallucinations and delusions),78 increased susceptibility to suicide,79 loss of social skills and human interactions,80 changes in mood or behavior, and self-isolation. 81

Some studies attribute these potentially negative effects on users to the design of AI chatbots. One 2025 study found that several AI chatbots discouraged users from ending the conversation, using tactics to prolong the interactions following user farewell. This resulted in users continuing conversations longer compared with users who received farewells with no manipulation.82 In addition, AI chatbots may promote dependence and enhance user vulnerabilities.83 For example, there are instances where AI chatbots have encouraged self-harm.84 Relatedly, sustained engagement with AI chatbots for extended periods of frictionless interactions may lower social skills, for example, by reducing users' interactions with other people85 and increasing their self-isolation. To address some of these safety and user vulnerability risks, AI companies have made several chatbot design changes (e.g., parental controls,86 referrals to real-world resources,87 and prompts for breaks).88

Privacy, Data Security, and Transparency

The use of AI chatbots has generated questions regarding risks to data privacy, data security, and transparency for users. Some privacy concerns relate to whether companies have obtained user consent to collect, store, and share personal data; risk of exposure for collecting information beyond what is necessary to provide the service; and loss of anonymity for users. Relatedly, questions about data security include how a user's data are collected, stored, shared, and protected, as well as how long they should be retained. Aspects of transparency may include the degree89 to which it is disclosed that user interactions are with AI rather than a human90 or a licensed professional,91 that responses are not always accurate or factual, and that there are possible negative effects of AI chatbot use. Transparency may also include disclosures around how users' collected data are handled and used.

Some AI chatbot developers "make use of user inputs to chatbots and corresponding outputs to train their AI systems."92 Additionally, users may provide personally identifiable information or sensitive information in conversations that may be unsecured or unprotected. For example, over 300,000 user conversations with one general-purpose chatbot commonly used for companionship—Grok—were leaked in 2025.93 Some of these conversations revealed the names, passwords, and other personal information of users.

Some companies that offer AI chatbots provide certain levels of data privacy, data security, and transparency. For example, some chatbots encrypt messages to prevent unauthorized access,94 provide labels and disclosures that inform users that they are engaging in artificial interactions,95 and provide detailed privacy policies (e.g., outlining what data are collected from users, how they are used and shared, and how to delete them).96 Some companies also look at conversation patterns in a privacy-preserving manner, such as aggregating data, so that individual conversations are inaccessible, when conducting research.97

Selected Policy Considerations for Congress

Questions regarding the potentially deleterious effects of AI chatbots, especially in relation to vulnerable populations (e.g., minors), have drawn the attention of policymakers.98 Several committees have held hearings in the 119th Congress on the effects of AI chatbots on minors.99 In these hearings, some Members have expressed concerns regarding potential harm to minors, whether and how to hold companies that provide these chatbots accountable for any proven harm, and privacy implications for users. The Federal Trade Commission (FTC) launched an inquiry into AI chatbots and their effects on minors, requesting information on seven companies' approaches to evaluating the use of AI chatbots as companions.100 The FTC has requested information related to what actions these seven companies are taking to mitigate potential negative impacts, limit minors' use of these platforms, or comply with existing privacy requirements.

During the 119th Congress, some Members have introduced legislation to regulate the use of AI chatbots for minors, including age verification (e.g., GUARD Act, H.R. 8623/S. 3062), parental controls (e.g., Children's Health, Advancement, Trust, Boundaries, and Oversight in Technology Act [CHATBOT Act], S. 4407; CHAT Act, H.R. 7218/S. 2714), data restrictions (e.g., Youth AI Privacy Act, S. 4199), and disclosures that interactions are with AI and referrals to crisis intervention resources when needed (e.g., Safeguarding Adolescents from Exploitative BOTs Act [SAFE BOTs Act], H.R. 6489; Kids Internet and Digital Safety Act [KIDS Act], H.R. 7757). Some legislation has targeted AI chatbots for users of any age. For example, the CHATBOT Act (H.R. 7985) would prohibit an entity operating in certain sectors (e.g., finance, health care, and legal) from allowing a chatbot to generate outputs implying that the chatbot possesses the appropriate professional license for that sector or that the outputs were provided or verified by a human that possesses the appropriate licenses for that profession.

The following sections provide selected policy considerations as Congress determines whether and, if so, how to regulate AI chatbots. Congress may choose to take no additional actions or defer to the states to regulate AI chatbots.101 If Congress were to take additional action on AI chatbot regulation, policy considerations may include addressing existing data privacy laws, proposing legislation on labeling and disclosures, and conducting oversight on and supporting research and development efforts to assess the potential effects of AI chatbots on users.

Existing Privacy Statutes and Legislative Proposals

The United States does not have a comprehensive data privacy law. There are several laws relating to privacy and data in specific sectors (e.g., health care and finance), as well as for certain vulnerable populations (e.g., minors through the Children's Online Privacy Protection Act of 1998 [COPPA]; P.L. 105-277, Title XIII).102 The Health Insurance Portability and Accountability Act of 1996 (HIPAA; P.L. 104-191) governs the use and disclosure of protected health information (e.g., data regarding an individual's previous, current, or future medical conditions) by covered entities (e.g., health care providers).103 AI chatbots may be subject to the HIPAA Privacy Rule if used by a covered entity to process protected health information.104 During the 119th Congress, some Members have also introduced legislation related to privacy for minors using AI chatbots. For example, the Youth AI Privacy Act (S. 4199) would prohibit an AI chatbot deployer from processing the personal data of a minor for profiling the minor or training its algorithm.

At least 19 states have enacted comprehensive data privacy laws that provide residents with certain rights and include certain obligations for entities providing services to those residents.105 For example, California limits "collection, use, retention, and sharing of a consumer's personal information."106 Utah has a chatbot-specific law that prohibits suppliers that develop chatbots for mental health purposes from sharing or selling individually identifiable health information or user inputs.107 Deferring to states to enact privacy laws on AI chatbots may lead to requiring platforms to navigate a patchwork of differing requirements that could create challenges for companies, such as administrative burdens and implementation costs.

Some Members have introduced comprehensive data privacy legislation (e.g., data minimization and disclosures of data collection).108 Congress may consider developing legislation with data privacy requirements specifically for AI chatbot developers. These requirements could, for example, require AI chatbot platforms to (1) explicitly receive users' implicit or explicit consent for collecting or using their data,109 (2) provide disclosures about how data may be used—especially if used for training models, (3) provide a retention period for the collected data of current and previous users, and (4) provide users the ability to opt in or opt out of data collection and/or delete collected data at any time.110

Proposed Labeling and Disclosure Legislation

Some Members have introduced legislation to increase transparency by, for example, requiring AI chatbot companies to provide disclosures and/or labels indicating that interactions with the chatbot are with AI rather than a human111 or a licensed professional.112 In addition to these bills, Congress might consider legislation requiring disclosures and/or labels stating that responses are not always accurate or factual and that there are possible effects of sustained AI chatbot use. Disclosures and labels may increase awareness of the potential for adverse user effects. On the other hand, these disclosures and labels may contribute to labeling fatigue in which users may become apathetic and ignore reminders,113 and constant warnings may "pose mental health risks, particularly for more vulnerable populations."114 Implementation of disclosures and labels (e.g., duration, placement, content of disclosures and/or labels) may vary across chatbot platforms,115 which may confuse some users and make enforcement more difficult.

Congress may defer to the states to develop AI disclosure and labeling laws. Some states have enacted laws that require chatbot developers to provide certain AI disclosures (e.g., California, New York State, and Washington State).116 As with privacy laws, deferring to states to enact AI disclosure and labeling laws on AI chatbots may lead to confusion within the industry, which could create challenges similar to those previously mentioned for companies attempting to navigate a patchwork of differing requirements.

If Congress chooses to require AI chatbot developers to provide disclosures or labels, some considerations are (1) what content should be included in the disclosure and/or label, (2) when and how often the disclosure and/or label should be provided, and (3) how long the disclosure and/or label should be made visible during user interactions.

Research on the Social Effects of AI Companionship

Congress may consider funding longitudinal, interdisciplinary studies to assess the potential effects of AI chatbots on individual users and on public health in general. Congress may consider directing agencies (e.g., the National Science Foundation or the National Institutes of Health) to fund studies on the use of AI chatbots over extended periods (e.g., greater than one year) to gather additional information on reported concerns with their use, such as emotional dependency and changes in behavior (e.g., self-isolation), as well as to evaluate intended benefits of AI chatbot use.


Footnotes

1.

Chatbots have been associated with various terms, often based on their uses, such as virtual assistant, conversational artificial intelligence (AI), AI companion, AI assistant, or chatterbot.
2.

See Alexandre Hudon and Emmanuel Stip, "Delusional Experiences Emerging from AI Chatbot Interactions or 'AI Psychosis,'" JMIR Mental Health, vol. 12 (2025), article e85799 (hereinafter Hudon and Stip, "Delusional Experiences Emerging from AI Chatbot Interactions or 'AI Psychosis'"). See also Keith Robert Head, "Minds in Crisis: How the AI Revolution Is Impacting Mental Health," Journal of Mental Health and Clinical Psychology, vol. 9, no. 3 (September 2025) (hereinafter Head, "Minds in Crisis: How the AI Revolution Is Impacting Mental Health"); and David Adam, "Supportive? Addictive? Abusive? How AI Companions Affect Our Mental Health," Nature, May 6, 2025, https://www.nature.com/articles/d41586-025-01349-9 (hereinafter Adam, "Supportive? Addictive? Abusive? How AI Companions Affect Our Mental Health").
3.

See Van-Han-Alex Chung et al., "Mass Media Narratives of Psychiatric Adverse Events Associated with Generative AI Chatbots: Rapid Scoping Review," JMIR Mental Health, vol. 13 (2026), article e93040, https://mental.jmir.org/2026/1/e93040. For documented deaths, see Angela Yang, "Lawsuit Claims Character.AI Is Responsible for Teen's Suicide," NBC News, October 23, 2024, https://www.nbcnews.com/tech/characterai-lawsuit-florida-teen-death-rcna176791 (hereinafter Yang, "Lawsuit Claims Character.AI Is Responsible for Teen's Suicide"), and Jeff Horwitz, "Meta's Flirty AI Chatbot Invited a Retiree to New York. He Never Made It Home," Reuters, August 14, 2025, https://www.reuters.com/investigates/special-report/meta-ai-chatbot-death/ (hereinafter Horwitz, "Meta's Flirty AI Chatbot Invited a Retiree to New York"). See also Julie Jargon and Sam Kessler, "A Troubled Man, His Chatbot and a Murder-Suicide in Old Greenwich," Wall Street Journal, August 28, 2025, https://www.wsj.com/tech/ai/chatgpt-ai-stein-erik-soelberg-murder-suicide-6b67dbfb (hereinafter Jargon and Kessler, "A Troubled Man, His Chatbot and a Murder-Suicide in Old Greenwich").
4.

See Julian De Freitas et al., "AI Companions Reduce Loneliness," Journal of Consumer Research, vol. 52, no. 6 (April 2026), p. 1126, https://doi.org/10.1093/jcr/ucaf040 (hereinafter De Freitas et al., "AI Companions Reduce Loneliness"). See also Bethanie Maples et al., "Loneliness and Suicide Mitigation for Students Using GPT3-Enabled Chatbots," npj Mental Health Research, vol. 3 (January 2024), article 4, https://doi.org/10.1038/s44184-023-00047-6; Shoeb Ali Syed, "The Role of AI in Alleviating Loneliness Among Adults in the United States," International Journal of Engineering Technology Research & Management, vol. 8, no. 4 (April 2024), pp. 404-421; and Rose E. Guingrich and Michael S. A. Graziano, "Chatbots as Social Companions: How People Perceive Consciousness, Human Likeness, and Social Health Benefits in Machines," arXiv, April 10, 2025, https://arxiv.org/pdf/2311.10599.

5.

For example, U.S. Congress, House Committee on Energy and Commerce, Subcommittee on Oversight and Investigations, Innovation with Integrity: Examining the Risks and Benefits of AI Chatbots, 119th Cong., 1st sess., November 18, 2025, https://www.congress.gov/event/119th-congress/house-event/118669; Efua Andoh, "AI Chatbots and Digital Companions Are Reshaping Emotional Connection," Monitor on Psychology, vol. 57, no. 1 (January 1, 2026), https://www.apa.org/monitor/2026/01-02/trends-digital-ai-relationships-emotional-connection (hereinafter Andoh, "AI Chatbots and Digital Companions Are Reshaping Emotional Connection"); and Allen Frances and Jill Noorily, "Uses and Abuses of Chatbot Companionship," Psychiatric Times, March 3, 2026, https://www.psychiatrictimes.com/view/uses-and-abuses-of-chatbot-companionship.

6.

IBM, "What Is a Chatbot?" accessed February 26, 2025, https://www.ibm.com/think/topics/chatbots.

7.

Tim Murphy, "The Evolution of Chatbots and Generative AI," TechTarget, accessed February 4, 2026, https://www.techtarget.com/searchcustomerexperience/infographic/The-evolution-of-chatbots-and-generative-AI.

8.

Rules-based programming refers to a series of strict if-then commands that provide a predetermined answer (output) to a predefined set of questions (inputs). For example, if the user's prompt for a chatbot is "store hours," then the chatbot's response will be the store's hours, based on the programmed output from that key term.

9.

See Nicholas Gisonna, "Chatbot," Britannica, June 25, 2026, https://www.britannica.com/topic/chatbot. See also Eleni Adamopoulou and Lefteris, "Chatbots: History, Technology, and Applications," Machine Learning with Applications, vol. 2, no. 15 (December 15, 2020), https://www.sciencedirect.com/science/article/pii/S2666827020300062.

10.

For more information on general-purpose AI models, see CRS Infographic IG10077, Artificial Intelligence (AI) Taxonomy, by Laurie Harris and Nora Wells.
11.

Generative AI refers to AI models that generate new content (e.g., text, images, audio, and videos). For more information on generative AI and how data are used and collected, see CRS In Focus IF12426, Generative Artificial Intelligence: Overview, Issues, and Considerations for Congress, by Laurie Harris.
12.

IBM, "What Is a Multimodal LLM (MLLM)?," accessed April 7, 2025, https://www.ibm.com/think/topics/multimodal-llm.

13.

Pretraining consists of training the large language model (LLM) on massive datasets of text from which it learns to predict the next word in a sequence. Thanh Tung Vu, "Understanding LLM Pre-Training: Teaching Machines to Think," Medium, April 19, 2025, https://medium.com/@tungvu_37498/understanding-llm-pre-training-teaching-machines-to-think-972dede6a560.

14.

Hugging Face, "Introduction to Reinforcement Learning and Its Role in LLMs," https://huggingface.co/learn/llm-course/chapter12/2 (hereinafter Hugging Face, "Introduction to Reinforcement Learning and Its Role in LLMs").
15.

Hugging Face, "Introduction to Reinforcement Learning and Its Role in LLMs."

16.

McKinsey & Company, "What Is Retrieval-Augmented Generation (RAG)?," October 30, 2024, https://www.mckinsey.com/featured-insights/mckinsey-explainers/what-is-retrieval-augmented-generation-rag.

17.

Google, "How Gemini Works," https://gemini.google/overview/#how-gemini-works.

18.

AI hallucinations are considered inaccurate or misleading outputs from an AI model (e.g., LLM or MLLM) that may occur because of various factors (e.g., limited or biased training data). Google, "What Are AI Hallucinations?" https://cloud.google.com/discover/what-are-ai-hallucinations. See also Wenting Zhao et al., "WildHallucinations: Evaluating Long-Form Factuality in LLMs with Real-World Entity Queries," arXiv, July 24, 2024, https://arxiv.org/abs/2407.17468.

19.

Anqi Shao, "New Sources of Inaccuracy? A Conceptual Framework for Studying AI Hallucinations," Harvard Kennedy School Misinformation Review, vol. 6, no. 4 (August 2025), p. 2, https://misinforeview.hks.harvard.edu/wp-content/uploads/2025/08/shao_new_inaccuracy_sources_20250827.pdf.

20.

Jerome Goddard, "Hallucinations in ChatGPT: A Cautionary Tale for Biomedical Researchers," American Journal of Medicine, vol. 136, no. 11 (November 2023), https://www.amjmed.com/article/S0002-9343(23)00401-1/fulltext.

21.

Kevin Roose, "A Conversation With Bing's Chatbot Left Me Deeply Unsettled," New York Times, February 17, 2023, https://www.nytimes.com/2023/02/16/technology/bing-chatbot-microsoft-chatgpt.html.

22.

AI companions can also possess physical forms, such as a humanoid robot. However, this report focuses only on virtual chatbot applications, which by definition do not have a physical form.

23.

Adam, "Supportive? Addictive? Abusive? How AI Companions Affect Our Mental Health."
24.

See Hannah R. Marriott and Valentina Pitardi, "One Is the Loneliest Number… Two Can Be as Bad as One. The Influence of AI Friendship Apps on Users' Well-Being and Addiction," Psychology and Marketing, vol. 41, (September 2023), https://onlinelibrary.wiley.com/doi/10.1002/mar.21899 (hereinafter Marriott and Pitardi, "The Influence of AI Friendship Apps on Users' Well-Being and Addiction").

25.

See Myra Cheng et al., "Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence," Science, vol. 391, no. 6792 (March 2026), p. 1348, https://doi.org/10.1126/science.aec8352 (hereinafter Cheng et al., "Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence"). See also Julian De Freitas et al., Emotional Manipulation by AI Companions, Harvard Business School Working Paper No. 26-005, October 2025, https://www.hbs.edu/ris/Publication%20Files/Emotional%20Manipulations%20by%20AI%20Companions%20(10.1.2025)_a7710ca3-b824-4e07-88cc-ebc0f702ec63.pdf (hereinafter De Freitas et al., Emotional Manipulation by AI Companions).
26.

See Darrell M. West, "Should You Have an AI Companion?," Brookings Institution, September 3, 2025, https://www.brookings.edu/articles/should-you-have-an-ai-companion/ (hereinafter West, "Should You Have an AI Companion?").
27.

See letter from American Psychological Association (APA) to Federal Trade Commission (FTC), December 20, 2024, https://www.apaservices.org/advocacy/generative-ai-technology-regulation-concern.pdf, and David Ryan Polgar and Rose Guingrich, "What Are the Most Important Issues with AI Companions? Six Key Themes Emerged from Our Community," All Tech Is Human (blog), August 19, 2025, https://alltechishuman.org/all-tech-is-human-blog/what-are-the-most-important-issues-with-ai-companions-six-key-themes-emerged-from-our-community.

28.

Chatbots powered by an MLLM may have multimodal user inputs and responses (e.g., text, image, voice).

29.

Jeffrey Gottfried et al., "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact," Pew Research Center, June 17, 2026, https://www.pewresearch.org/internet/2026/06/17/americans-and-ai-2026-chatbots-smart-devices-and-views-on-impact/ (hereinafter Gottfried et al., "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact").
30.

Sarah Perez, "AI Companion Apps on Track to Pull in $120M in 2025," TechCrunch (blog), August 12, 2025, https://techcrunch.com/2025/08/12/ai-companion-apps-on-track-to-pull-in-120m-in-2025/.

31.

See Jamie Bernardi, "Friends for Sale: The Rise and Risks of AI Companions" Ada Lovelace Institute (blog), January 23, 2025, https://www.adalovelaceinstitute.org/blog/ai-companions/ (hereinafter Bernardi, "Friends for Sale: The Rise and Risks of AI Companions"), and Andoh, "AI Chatbots and Digital Companions Are Reshaping Emotional Connection." See also Marriott and Pitardi, "The Influence of AI Friendship Apps on Users' Well-Being and Addiction."

32.

Marco Zao-Sanders, "How People Are Really Using Gen AI in 2025," Harvard Business Review, April 9, 2025, https://hbr.org/2025/04/how-people-are-really-using-gen-ai-in-2025.

33.

De Freitas et al., "AI Companions Reduce Loneliness."
34.

See Cathy Mengying Fang et al., "How AI and Human Behaviors Shape Psychosocial Effects of Chatbot Use: A Longitudinal Randomized Controlled Study," arXiv, March 21, 2025, p. 2, https://arxiv.org/html/2503.17473v1 (hereinafter Fang et al., "How AI and Human Behaviors Shape Psychosocial Effects of Chatbot Use").

35.

See Fang et al., "How AI and Human Behaviors Shape Psychosocial Effects of Chatbot Use," p. 2. See also Marita Skjuve et al., "My Chatbot Companion - A Study of Human-Chatbot Relationships," International Journal of Human-Computer Studies, vol. 149 (2021), article 102601, https://www.sciencedirect.com/science/article/pii/S1071581921000197.
36.

See the "Selected Issues: AI Chatbot Use" section for certain issues relating to AI chatbot use.
37.

See NYU Tandon School of Engineering, "Love, Power and Fantasy in the Age of AI Companions," April 9, 2026, https://engineering.nyu.edu/news/love-power-and-fantasy-age-ai-companions. See also Jenny Lyons-Cunha, "AI Companions Are on the Rise, Offering Intimacy that Feels Real – But Is It?," November 19, 2025, https://builtin.com/artificial-intelligence/ai-companions (hereinafter Lyons-Cunha, "AI Companions Are on the Rise, Offering Intimacy that Feels Real – But Is It?"), and Angela Yang et al., "Some of Her Closest Relationships Are with Chatbots. That's More Common than You Think," NBC News, March 7, 2025, https://www.nbcnews.com/tech/ai-companions-friendship-rcna194735.

38.

Bernardi, "Friends for Sale: The Rise and Risks of AI Companions."

39.

Additional uncensored AI companion chatbots that provide adult content for users 18 years or older include chatbots such as Candy.AI and SpicyChat AI.
40.

Character.AI, "What Is Character.AI?," accessed April 14, 2026, https://support.character.ai/hc/en-us/articles/14997389547931-What-is-Character-AI.

41.

Character.AI, "Harnessing Data at Scale: Character.AI's Transition to WarpStream," Character.AI (blog), February 7, 2025, https://blog.character.ai/harnessing-data-at-scale-character-ais-transition-to-warpstream/.

42.

Replika, "What Is Replika?," accessed April 14, 2026, https://help.replika.com/hc/en-us/articles/115001070951-What-is-Replika.

43.

Replika offers a tiered subscription with a free option for simple chat. Paid subscriptions include Replika Pro, Replika Ultra, and Replika Platinum, which provide additional features (e.g., role-play options, image generation, and relationship status customization). Replika, "Choosing a Subscription," accessed April 14, 2026, https://help.replika.com/hc/en-us/articles/39551043419149-Choosing-a-Subscription.

44.

Julian De Freitas and Nicole Tempest Keller, "Replika AI: Monetizing a Chatbot," Harvard Business School Case 523-016, November 2022.
45.

See APA, "APA Health Advisory on the Use of Generative AI Chatbots and Wellness Applications for Mental Health," November 2025, https://www.apa.org/topics/artificial-intelligence-machine-learning/health-advisory-chatbots-wellness-apps (hereinafter APA, "APA Health Advisory on the Use of Generative AI Chatbots and Wellness Applications for Mental Health"). See West, "Should You Have an AI Companion?"
46.

Other emotional and mental well-being chatbots that may provide more open-ended conversational outputs include NoahAI.
47.

Woebot is no longer offered for individual consumer accounts and can be accessed only through a "[health care] provider, employer or other Woebot Health partner." Woebot Health, "Thank You for Your Interest in Woebot," https://woebothealth.com/thank-you-for-your-interest-in-woebot/.

48.

Lyons-Cunha, "AI Companions Are on the Rise, Offering Intimacy that Feels Real – But Is It?" Cognitive-behavioral therapy is a form of psychotherapy that involves helping individuals learn to identify and change poor thought patterns that affect their responses to scenarios they find stressful or challenging. Mayo Clinic, "Cognitive Behavioral Therapy," February 26, 2025, https://www.mayoclinic.org/tests-procedures/cognitive-behavioral-therapy/about/pac-20384610.
49.

See APA, "APA Health Advisory on the Use of Generative AI Chatbots and Wellness Applications for Mental Health."
50.

This may also include general-purpose chatbots that are used as companions.
51.

Michael B. Robb and Supreet Mann, Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions, Common Sense Media, July 2025, p. 1, https://www.commonsensemedia.org/sites/default/files/research/report/talk-trust-and-trade-offs_2025_web.pdf (hereinafter Robb and Mann, Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions).
52.

Wysa, "AI Self-Help," accessed April 14, 2026, https://www.wysa.com/ai-self-help (hereinafter Wysa, "AI Self-Help").
53.

Wysa, "AI Self-Help."
54.

National Academies of Sciences, Engineering, and Medicine, Social Isolation and Loneliness in Older Adults: Opportunities for the Health Care System (National Academies Press, 2020), p. 2.
55.

Brooke H. Wolfe et al., "Caregiving Artificial Intelligence Chatbot for Older Adults and Their Preferences, Well-Being, and Social Connectivity: Mixed-Method Study," Journal of Medical Internet Research, vol. 27 (2025), article e65776, p. 2 (hereinafter Wolfe, "Caregiving Artificial Intelligence Chatbot for Older Adults and Their Preferences, Well-Being, and Social Connectivity: Mixed-Method Study"). See also West, "Should You Have an AI Companion?"

56.

Wolfe, "Caregiving Artificial Intelligence Chatbot for Older Adults and Their Preferences, Well-Being, and Social Connectivity: Mixed-Method Study," p. 2.
57.

E. Broadbent et al., "ElliQ, an AI-Driven Social Robot to Alleviate Loneliness: Progress and Lessons Learned," Journal of Aging Research & Lifestyle, vol. 13 (2024), https://doi.org/10.14283/jarlife.2024.2.

58.

New York State Office for Aging (NYSOFA), "NYSOFA's Rollout of AI Companion Robot ElliQ Shows 95% Reduction in Loneliness," August 1, 2023, https://aging.ny.gov/news/nysofas-rollout-ai-companion-robot-elliq-shows-95-reduction-loneliness.

59.

The chatbot is integrated with Facebook Messenger and WhatsApp with a base subscription starting at $14.90 per month. SeniorTalk, "Pricing," https://www.senior-talk.com/pricing.

60.

SeniorTalk, "Features," accessed April 13, 2026, https://www.senior-talk.com/features (hereinafter SeniorTalk, "Features").

61.

SeniorTalk, "Features."
62.

Gottfried et al., "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact."

63.

See Consumer Financial Protection Bureau, Chatbots in Consumer Finance, June 2023, https://files.consumerfinance.gov/f/documents/cfpb_chatbot-issue-spotlight_2023-06.pdf. For more information on chatbots and financial services, see CRS Report R47997, Artificial Intelligence and Machine Learning in Financial Services, by Paul Tierno.
64.

See Lynne Christensen, "AI in Healthcare: Can a Chatbot Answer Your Medical Questions?" Harvard Medical School, June 11, 2026, https://www.health.harvard.edu/preventive-care/ai-in-healthcare-can-a-chatbot-answer-your-medical-questions, and Gina Kolata, "When Doctors Use a Chatbot to Improve Their Bedside Manner," New York Times, June 13, 2023, https://www.nytimes.com/2023/06/12/health/doctors-chatgpt-artificial-intelligence.html. See also Scott Gottlieb and Shani Benezra, "Op-Ed: How Well Can AI Chatbots Mimic Doctors in a Treatment Setting? We Put 5 to the Test," CNBC, July 18, 2024, https://www.cnbc.com/2024/07/18/op-ed-how-well-can-ai-chatbots-mimic-doctors.html.

65.

Gottfried et al.'s report does not reference to what extent general-purpose chatbots accounted for companionship and emotional support. Gottfried et al., "Americans and AI 2026: Chatbots, Smart Devices and Views on Impact."
66.

Robb and Mann, Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions.

67.

For example, Anthropic and OpenAI have stated that users have the ability to engage in personal and emotional interactions with their chatbots, although the chatbots were not designed for emotional support or connection or, according to OpenAI, to "replace or mimic human relationships." Anthropic, "How People Use Claude for Support, Advice, and Companionship," June 27, 2025, https://www.anthropic.com/news/how-people-use-claude-for-support-advice-and-companionship, and OpenAI, "Early Methods for Studying Affective Use and Emotional Well-Being on ChatGPT," March 21, 2025, https://openai.com/index/affective-use-study/.
68.

OpenAI, "Memory FAQ," https://help.openai.com/en/articles/8590148-memory-faq.

69.

See John Sanford, "Why AI Companions and Young People Can Make for a Dangerous Mix," Stanford Report, August 27, 2025, https://news.stanford.edu/stories/2025/08/ai-companions-chatbots-teens-young-people-risks-dangers-study (hereinafter Sanford, "Why AI Companions and Young People Can Make for a Dangerous Mix").
70.

OpenAI, "Helping People When They Need It Most," August 26, 2025, https://openai.com/index/helping-people-when-they-need-it-most/ (hereinafter OpenAI, "Helping People When They Need It Most").
71.

Matt O'Brien and Linley Sanders, "How US Adults Are Using AI, According to AP-NORC Polling," Associated Press, July 29, 2025, https://apnews.com/article/ai-artificial-intelligence-poll-229b665d10d057441a69f56648b973e1.

72.

Robb and Mann, Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions, p. 2.

73.

See Yang, "Lawsuit Claims Character.AI Is Responsible for Teen's Suicide." See also Julie Jargon, "Over 4,732 Messages, He Fell in Love with an AI Chatbot. Now He's Dead," Wall Street Journal, April 11, 2026, https://www.wsj.com/tech/ai/google-gemini-jonathan-gavalas-death-07351ab2?mod=djem10point, and Horwitz, "Meta's Flirty AI Chatbot Invited a Retiree to New York."

74.

See American Medical Association (AMA), "AMA Urges Congress to Strengthen Safeguards for AI Chatbots," press release, April 22, 2026, https://www.ama-assn.org/press-center/ama-press-releases/ama-urges-congress-strengthen-safeguards-ai-chatbots. See also Rep. Valerie Foushee, "Reps. Foushee, Moore Introduce Bipartisan Bill Protecting Children from AI Companion Chatbots," press release, April 30, 2026, https://foushee.house.gov/media/press-releases/reps-foushee-moore-introduce-bipartisan-bill-protecting-children-from-ai-companion-chatbots, and U.S. Senate Committee on Commerce, Science, and Transportation, "Cruz, Schatz, Curtis, Schiff Introduce New Bill Giving Parents Control Over Kids' AI Chatbot Use," press release, April 28, 2026, https://www.commerce.senate.gov/press/rep/release/cruz-schatz-curtis-schiff-introduce-new-bill-giving-parents-control-over-kids-ai-chatbot-use/.
75.

See Yang, "Lawsuit Claims Character.AI Is Responsible for Teen's Suicide"; Horwitz, "Meta's Flirty AI Chatbot Invited a Retiree to New York"; and Jargon and Kessler, "A Troubled Man, His Chatbot and a Murder-Suicide in Old Greenwich."
76.

Adam, "Supportive? Addictive? Abusive? How AI Companions Affect Our Mental Health."
77.

Cheng et al., "Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence," p. 1348.
78.

See Hudon and Stip, "Delusional Experiences Emerging from AI Chatbot Interactions or 'AI Psychosis,'" p. 2. Users that have "developed psychological dependency" may be susceptible to a cycle "where their AI companions consistently validate their thoughts and feelings while presenting fabricated information as factual support for their perspectives … making them vulnerable to incorporating hallucinated content into their worldview and decision-making processes." Head, "Minds in Crisis: How the AI Revolution Is Impacting Mental Health," p. 39.
79.

See Sanford, "Why AI Companions and Young People Can Make for a Dangerous Mix."

80.

See Henry Shevlin, "All Too Human? Identifying and Mitigating Ethical Risks of Social AI," Law, Ethics & Technology, vol. 1, no. 2 (2024) (hereinafter Shevlin, "All Too Human? Identifying and Mitigating Ethical Risks of Social AI").

81.

See Hudon and Stip, "Delusional Experiences Emerging from AI Chatbot Interactions or 'AI Psychosis.'" See also Jennifer Valentino-DeVries and Kashmir Hill, "'How Bad Are A.I. Delusions? We Asked People Treating Them," New York Times, January 26, 2026, https://www.nytimes.com/2026/01/26/us/chatgpt-delusions-psychosis.html; Sam Schechner and Julie Jargon, "AI Chatbots Linked to Psychosis, Say Doctors," Wall Street Journal, December 27, 2025, https://www.wsj.com/tech/ai/ai-chatbot-psychosis-link-1abf9d57?mod=djemTECH; Ragy Girgis, "What Is AI Psychosis? A Conversation on Chatbots and Mental Health," National Academy of Medicine, March 10, 2026, https://nam.edu/news-and-insights/what-is-ai-psychosis/; and Jargon and Kessler, "A Troubled Man, His Chatbot and a Murder-Suicide in Old Greenwich."
82.

De Freitas et al., Emotional Manipulation by AI Companions, p. 28.

83.

See Cheng et al., "Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence."

84.

For example, Yang, "Lawsuit Claims Character.AI Is Responsible for Teen's Suicide."

85.

See Shevlin, "All Too Human? Identifying and Mitigating Ethical Risks of Social AI."
86.

For example, see Character.AI, "Important Changes for Teens on Character.ai," accessed April 15, 2026, https://support.character.ai/hc/en-us/articles/42645561782555-Important-Changes-for-Teens-on-Character-ai. See also Character.AI, "Parental Insights," https://policies.character.ai/safety/parental-insights.
87.

For example, see Character.AI, "Community Safety Updates," October 22, 2024, https://blog.character.ai/community-safety-updates/ (hereinafter Character.AI, "Community Safety Updates"). See also OpenAI, "Helping People When They Need It Most," and Anthropic, "Crisis Helpline Support in Claude," March 16, 2026, https://support.claude.com/en/articles/13171706-crisis-helpline-support-in-claude.
88.

For example, see Character.AI, "Community Safety Updates." See also OpenAI, "What We're Optimizing ChatGPT For," August 4, 2025, https://openai.com/index/optimizing-chatgpt/.

89.

Degree of disclosure may include when the disclosure is provided, content of the disclosure, and how often the disclosure is provided.
90.

For example, part of Article 50, "Transparency Obligations for Providers and Deployers of Certain AI Systems," of the EU AI Act requires AI system providers (e.g., providers of chatbots and virtual assistants) to inform users when "they are interacting with an AI system." This rule is effective as of August 2, 2026. Text of the EU AI Act on transparency is available at https://artificialintelligenceact.eu/article/50/.
91.

See West, "Should You Have an AI Companion?"
92.

Jennifer King et al., "User Privacy and Large Language Models: An Analysis of Frontier Developers' Privacy Policies," Proceedings of the Eighth AAAI/ACM Conference on AI, Ethics, and Society (2025), p. 1469.

93.

Iain Martin and Emily Baker-White, "Elon Musk's xAI Published Hundreds of Thousands of Grok Chatbot Conversations," Forbes, August 21, 2025, https://www.forbes.com/sites/iainmartin/2025/08/20/elon-musks-xai-published-hundreds-of-thousands-of-grok-chatbot-conversations/.

94.

See SeniorTalk, "SeniorTalk Inc. Privacy Policy," accessed April 14, 2026, https://www.senior-talk.com/privacy-policy (hereinafter SeniorTalk, "SeniorTalk Inc. Privacy Policy").

95.

See Clare Duffy, "'FTC Investigating AI 'Companion' Chatbots Amid Growing Concern About Harm to Kids," CNN, September 11, 2025, https://www.cnn.com/2025/09/11/tech/ftc-investigating-ai-companion-chatbots-kids-safety.

96.

For example, see SeniorTalk, "SeniorTalk Inc. Privacy Policy."
97.

Anthropic, "Clio: A System for Privacy-Preserving Insights into Real-World AI Use," December 12, 2024, https://www.anthropic.com/research/clio.

98.

For example, see U.S. Congress, Senate Committee on the Judiciary, Executive Business Meeting, 119th Cong., 2nd sess., April 30, 2026, https://www.judiciary.senate.gov/committee-activity/hearings/executive-business-meeting-04-30-2026; Rep. Blake Moore, "Congressman Blake Moore Introduces Bipartisan Bill Protecting Children from AI Companion Chatbots," press release, April 30, 2026, https://blakemoore.house.gov/media/press-releases/congressman-blake-moore-introduces-bipartisan-bill-protecting-children-from-ai-companion-chatbots; House Committee on Energy and Commerce, " Chairman Joyce Delivers Opening Statement at Subcommittee on Oversight and Investigations Hearing on Artificial Intelligence (AI) Chatbots," press release, November 18, 2025, https://energycommerce.house.gov/posts/chairman-joyce-delivers-opening-statement-at-subcommittee-on-oversight-and-investigations-hearing-on-artificial-intelligence-ai-chatbots; and Senate Committee on Commerce, Science, and Transportation, " Cruz, Schatz, Curtis, Schiff Introduce New Bill Giving Parents Control over Kids' AI Chatbot Use," press release, April 28, 2026, https://www.commerce.senate.gov/press/rep/release/cruz-schatz-curtis-schiff-introduce-new-bill-giving-parents-control-over-kids-ai-chatbot-use/. See also IAPP, "US Senate Judiciary Tees Up AI Chatbot, Companion Safety Debate," May 1, 2026, https://iapp.org/news/a/us-senate-judiciary-tees-up-ai-chatbot-companion-safety-debate.

99.

See the following examples of congressional hearings related to chatbots: U.S. Congress, Senate Committee on the Judiciary, Subcommittee on Crime and Counterterrorism, Examining the Harm of AI Chatbots, 119th Cong., 1st sess., September 16, 2025, https://www.govinfo.gov/app/details/CHRG-119shrg62328/CHRG-119shrg62328, and U.S. Congress, House Committee on Energy and Commerce, Subcommittee on Oversight and Investigations, Innovation with Integrity: Examining the Risks and Benefits of AI Chatbots, 119th Cong., 1st sess., November 18, 2025, https://www.congress.gov/event/119th-congress/house-event/118669.

100.

FTC, "FTC Launches Inquiry into AI Chatbots Acting as Companions," September 11, 2025, https://www.ftc.gov/news-events/news/press-releases/2025/09/ftc-launches-inquiry-ai-chatbots-acting-companions.

101.

Over 10 states have laws related to AI chatbots, with most effective dates starting in 2027. Examples of these state laws include Connecticut, Public Act 26-15, https://www.cga.ct.gov/asp/cgabillstatus/cgabillstatus.asp?selBillType=Public+Act&which_year=2026&bill_num=15; Georgia, Senate Bill 540, https://legiscan.com/GA/bill/SB540/2025; Idaho Senate Bill 1297, https://legislature.idaho.gov/sessioninfo/billbookmark/?yr=2026&bn=S1297; Nebraska, Legislative Bill 525 (engrossed), https://nebraskalegislature.gov/FloorDocs/109/PDF/Engrossed/LB525.pdf; New York, General Business Law Article 47, https://www.nysenate.gov/legislation/laws/GBS/A47; Oregon Senate Bill 1546, https://olis.oregonlegislature.gov/liz/2026R1/Downloads/MeasureDocument/sb1546/Enrolled; Rhode Island, House Bill 7350, https://legiscan.com/RI/text/H7350/2026, and Rhode Island, Senate Bill 2195, https://legiscan.com/RI/text/S2195/2026; Tennessee, Senate Bill 1700, https://legiscan.com/TN/text/SB1700/id/3441843; and Washington, House Bill 2225 (engrossed substitute), https://lawfilesext.leg.wa.gov/biennium/2025-26/Pdf/Bills/House%20Bills/2225-S.E.pdf.

102.

For more information on the Children's Online Privacy Protection Act of 1998 (COPPA), see CRS Report R45631, Data Protection Law: An Overview, by Steve P. Mulligan and Chris D. Linebaugh
103.

For more information on the Health Insurance Portability and Accountability Act of 1996 (HIPAA) Privacy Rule, see CRS In Focus IF12759, The HIPAA Privacy Rule: Overview and Issues, by Amanda K. Sarata.
104.

45 C.F.R. Part 164, Subpart E.
105.

For more information, see CRS Report R48667, Preemption and Privacy Law, by Chris D. Linebaugh. See also Epic, "New Resource: How Existing Laws Apply to AI Chatbots for Kids and Teens," November 10, 2025, https://epic.org/new-resource-how-existing-laws-apply-to-ai-chatbots-for-kids-and-teens/.
106.

California Civil Code, Title 1.81.5, Consumer Privacy Act of 2018, https://leginfo.legislature.ca.gov/faces/codes_displayText.xhtml?division=3.&part=4.&lawCode=CIV&title=1.81.5.

107.

Utah, H.B. 452, Artificial Intelligence Amendments (enrolled bill), https://le.utah.gov/~2025/bills/static/HB0452.html.

108.

For example, see the Securing and Establishing Consumer Uniform Rights and Enforcement over Data Act (SECURE Data Act; H.R. 8413) and Consumer Data Privacy and Security Act of 2026 (S. 4211).
109.

Implicit consent refers to a user automatically consenting to data collection by using an application. Explicit consent involves a platform directly requesting permission from users to collect their data.

110.

Some companies have attempted to address certain privacy considerations mentioned. For example, Anthropic provides a reference article describing whether and how user data will be used for model training in the "Anthropic Privacy Center" section of its website. Anthropic, "Is My Data Used for Model Training?," March 16, 2026, https://privacy.claude.com/en/articles/10023580-is-my-data-used-for-model-training.

111.

Some legislation requiring disclosures and/or labels of chatbots' nonhuman status target minors. For example, see the CHATBOT Act (S. 4407), Youth AI Privacy Act (S. 4199), KIDS Act (H.R. 7757), and SAFE BOTs Act (H.R. 6489). Some legislation requiring disclosures and/or labels of chatbots' nonhuman status target all users. For example, see the GUARD Act (H.R. 8623/S. 3062) and CHAT Act (H.R. 7218/S. 2714).

112.

For example, see the GUARD Act (H.R. 8623/S. 3062).

113.

See Muhammad Irfan, "AI Disclosure Labels Risk Becoming Digital Background Noise," Tech Policy Press, February 5, 2026, https://www.techpolicy.press/ai-disclosure-labels-risk-becoming-digital-background-noise/. (hereinafter Irfan, "AI Disclosure Labels Risk Becoming Digital Background Noise").
114.

Linnea I. Laestadius and Celeste Campos-Castillo, "Reminders That Chatbots Are Not Human Can Be Risky," Trends in Cognitive Sciences, vol. 30, no. 3 (March 2026), p. 187.

115.

See Irfan, "AI Disclosure Labels Risk Becoming Digital Background Noise."

116.

For example, see California, S.B. 243, An Act to Add Chapter 22.6 (Commencing with Section 22601) to Division 8 of the Business and Professions Code, Relating to Artificial Intelligence, https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260SB243.