(Maryland) H.B. 952: Companion Chatbots

Maryland General Assembly
Senate Finance Committee

Dear Chair Beidle, Vice Chair Hayes, and Members of the Committee:

EPIC supports the need for important legislation like H.B. 952, but we write to express significant concern over the way the bill is currently drafted. As you have heard already this session, chatbot use is increasing rapidly, and the harms this technology is causing are also skyrocketing. Policymakers should act swiftly to protect residents from the plethora of harms that chatbots are causing. However, as drafted, H.B. 952 is ineffective at solving these harms while also opening Maryland up to significant litigation risk. This bill should first be amended to ensure it is workable to implement and to decrease the risk that it is struck down in a First Amendment challenge. The bill should also be amended to ensure it is properly scoped to cover all chatbots that are causing harm and to add concrete protections for chatbot users and testing requirements for chatbot providers. Without these essential amendments, H.B. 952 is unworkable, ineffective, and poses a potentially expensive litigation risk to the State.

The Electronic Privacy Information Center (EPIC) is an independent nonprofit research organization in Washington, D.C., established in 1994 to protect privacy, freedom of expression, and democratic values in the information age.^[1] EPIC has a long history of advocating for safe and responsible technology and for tech policy that protects the privacy and civil rights of all people.

Chatbots are causing devastating harms to people of all ages.

Chatbots are quickly gaining traction among people of all ages. Two-thirds of teens report having used a chatbot, and 30% of teens use chatbots every day.[2] The number of adults who have used ChatGPT doubled from mid-2023 to mid-2025, with over one-third of adults having used the chatbot.[3] Use of chatbots is causing devastating harms, including suicide, self-harm, violence against others, sexual exploitation and predation, financial scams, reputational injuries, and mental health harms like anxiety, depression, unhealthy emotional attachment, and AI psychosis.[4] Because chatbots are almost entirely unregulated, these harms will only grow. The Legislature should act now to force tech companies to act more responsibly and design chatbots that are safer for everyone.

H.B. 952 requires significant amendments to be both workable and constitutional.

As it is currently drafted, the protections that H.B. 952 endeavors to provide to users are unclear and underdeveloped. For example, the bill is inconsistent in the terms used to refer to companies that operate a chatbot, and key terms are insufficiently defined or undefined altogether. The terms “developer,” “operator,” “chatbot provider,” and “controller” are used in the bill seemingly interchangeably, but only “operator” is defined. These terms need be made consistent—and a definition for “developer” added—for the bill to function correctly. Critically, the term “companion chatbot” is defined in a way that is underinclusive of chatbots that cause similar harm to minors and contains overbroad exemptions.

Additionally, key provisions of H.B. 952 are not properly tailored to the chatbot context. Chatbots function differently than other websites and online services. The data minimization provision in section (f) should be amended to the chatbot context to most effectively combat excessive data collection and mitigate data misuse. Specifically, (f)(2) should prohibit using data regarding emotional state or mental health vulnerabilities to tailor outputs from a chatbot to increase the duration or frequency of use. As it is currently written, the prohibition applies to using that data to tailor “algorithms,” which is vague in the chatbot context. Similarly, the complaint and takedown provisions in section (g) should be amended to better fit the chatbot context. Unlike social media platforms, chatbots do not host “content,” and a chatbot developer or operator cannot “take down” chatbot contents. Chatbots generate outputs that vary by user and context, and chatbot developers and operators can develop or refine software to prevent the chatbot from generating the output in the future. The complaint system in (g) should thus be tailored to regulate outputs from a chatbot, not “content” more broadly.

H.B. 952 should also be amended to reduce litigation risks and ensure compliance. Courts are still evaluating the extent to which chatbot regulation may impact speech rights of chatbot operators and users. To ensure that Marylanders benefit from the bill’s protections, the Committee should be careful to draft H.B. 952 with litigation risks in mind, considering potential First Amendment and vagueness challenges from the tech industry. For example, the entire bill is scoped based on a relatively narrow definition of companion chatbot with many exceptions. The definition should be amended and broadened to be more durable in court and more effective in practice. The chatbot industry is continuing to develop and expand, often in unanticipated settings, and this bill’s narrow definition of “companion chatbot” means that harms are likely to quickly outpace this regulation.

Further, regulating any chatbot outputs that “relate to” or “concern” certain topics is likely to be unconstitutionally overbroad. While First Amendment doctrine is unsettled in this area, the State is likely on the safest constitutional ground when it regulates demonstrably harmful chatbot outputs that a chatbot operator will likely not claim as their own speech, like those that encourage a user to commit suicide. The bill, as currently written, includes large swaths of outputs that chatbot operators would likely claim as their own expression, and whose regulation would likely trigger the highest First Amendment scrutiny. For example, outputs that “relate to” self-harm include outputs discouraging self-harm and outputs that provide resources for helping users expressing feelings of self-harm. Outputs “concerning” sexually explicit conduct include sex education materials and other information about sexually transmitted diseases, human sexuality, and safe sex. The Committee should, instead, amend the bill to regulate harmful outputs the chatbot operators would likely disclaim, such as outputs that “encourage” self-harm and suicide, that “affirm” a user’s expressed decision to self-harm or die by suicide, and that “direct or induce” a minor to engage in sexually explicit conduct. Formulating the safeguards this way accomplishes the original goal of the legislation—to keep chatbot users safer—without opening the State up to unnecessary First Amendment challenges.

Chatbot regulations also have to be carefully drafted to avoid vagueness challenges. There are a number of vague terms in the current version of the bill. For instance, a person of common intelligence would have to guess at what chatbot outputs are “likely to elicit emotional responses.” Every person reacts differently to information, and even the most mundane information can elicit an emotional response. A forecast of a sunny warm day is likely to elicit a happy response from many, a disappointed response from some (hoping for sweater weather, rain, or snow), and no response from others. The scope of “emotional responses” covered by the provision is also unclear, as the term is undefined, which contributes to both vagueness and overbreadth concerns. Similarly, it is unclear what it means for a chatbot to “be capable of meeting a user’s social needs,” as “social needs” is undefined. Many things could fall under this definition or not, and each user will have different “social needs.” Replacing these vague terms with more concrete ones will help the law pass constitutional muster and ensure that its important protections can be enforced.

As drafted, H.B. 952 does not do enough to provide meaningful protections to Marylanders. Additional provisions are necessary before H.B. 952 will effectively prevent harm.

In addition to amending H.B. 952 to make it workable and likely to withstand constitutional scrutiny, the bill should also be strengthened to more meaningfully protect Marylanders. As drafted, the bill requires companies operating chatbots to provide warnings to users that they are not interacting with a human and to establish a basic protocol for preventing companion chatbots from producing outputs that encourage users to harm themselves or that encourage minors to engage in sexually explicit behavior. While these safeguards are important steps toward making chatbots safer, they are far from sufficient to reduce the harms chatbots present.

H.B. 952 should be amended to also include restrictions on data-driven harms that are especially acute on chatbots, such as chatbot operators using personal data to tailor outputs to users to increase their engagement or target advertisements to users. Protections should also be added to ensure that companies do not advertise or make available chatbots that they claim can give qualified professional advice—like a therapist, doctor, lawyer, or accountant would—or that they will keep users’ information confidential. Finally, this bill should be amended to include requirements that chatbot operators regularly test these safeguards and protocols to ensure they are working to prevent harm.

Big Tech companies are currently operating with very few rules and very little oversight, and they’ve proven time and time again that they cannot be trusted to self-regulate. Thus, it is essential that the Committee advance a bill that sets clear rules of the road for chatbot providers to protect Marylanders from chatbot harms. As currently drafted, H.B. 952 does not achieve this goal, but with amendments, it could be an important step toward securing protections for Marylanders.

*          *          *

EPIC commends the Committee for working to solve the urgent harms chatbots are causing, but H.B. 952 must be significantly amended before it can address those harms. While we cannot afford to wait to act on this issue, passing a constitutionally vulnerable bill that is likely to be tied up in litigation for years does not protect Marylanders. EPIC urges the Committee to amend H.B. 952 before moving this important legislation forward. Thank you for the opportunity to testify. EPIC is happy to be a resource to the Committee on amending H.B. 952 to better protect Marylanders.

[1] EPIC, About EPIC, https://epic.org/about/.

[2] Michelle Faverio & Olivia Sidoti, Teens, Social Media and AI Chatbots 2025, Pew Research Ctr. (Dec. 9, 2025), https://www.pewresearch.org/internet/2025/12/09/teens-social-media-and-ai-chatbots-2025/.  

[3] Olivia Sidoti & Colleen McClain, 34% of U.S. Adults Have Used ChatGPT, About Double the Share in 2023, Pew Research Ctr. (June 25, 2025), https://www.pewresearch.org/short-reads/2025/06/25/34-of-us-adults-have-used-chatgpt-about-double-the-share-in-2023/.

[4] See, e.g., Kashmir Hill, They Asked an A.I. Chatbot Questions. The Answers Sent Them Spiraling., N.Y. Times (June 13, 2025), https://www.nytimes.com/2025/06/13/technology/chatgpt-ai-chatbots-conspiracies.html; Noor Al-Sibai, Psychiatrists Warn that Talking to AI Is Leading to Severe Mental Health Issues, Futurism (Aug. 19, 2025), https://futurism.com/psychiatrists-ai-mental-health-chatbots; Nguyen, Meyer & Levine, AI Sycophancy: Impacts, Harms & Questions, Georgetown Law Inst. for Technology Law & Policy (Aug. 11, 2025), https://www.law.georgetown.edu/tech-institute/research-insights/insights/ai-sycophancy-impacts-harms-questions/.