How My AI Companion Works For Seniors: Safety, Security, and Trust
If you’ve been researching an AI companion for seniors, aging parents, or yourself, your first question is probably: Is it safe? That’s the most important thing for all of us to consider in the modern world of the internet, but especially for older adults.
My AI Companion was built with this in mind. It is a phone-based service that calls and texts older adults on a schedule they choose. No app, no Wi-Fi, no new device to learn. This article walks through how it works, what the AI can and can’t do, and the safety and privacy protections sitting underneath every conversation that were developed with scientists and doctors from Dartmouth and Stanford and the Department of Veterans Affairs (VA).
Why AI Companions for Seniors?
AI companions can help reduce the harsh realities of social isolation, cognitive decline, and physical vulnerability to help seniors maintain independence in their own homes for as long as possible, also known as aging-in-place. In the past, the only solutions were expensive home-health aides or difficult transitions to assisted living, and seniors were at the mercy of their bank accounts for quality of later life.
A new ally has emerged: the My AI Companion for seniors. While artificial intelligence was once the stuff of science fiction, it is now becoming a fundamental pillar of senior care. However, as we embrace these digital friends, we must address the elephant in the room. While AI companions offer life-changing benefits for independence, true confidence in any product of this nature demands a safety-first design that protects personal data and physical well-being.
Do Seniors Really Need AI Companions?
The loneliness epidemic among seniors is more than just a feeling; it is a clinical risk factor. Research from the National Academies of Sciences, Engineering, and Medicine has linked social isolation to a significantly increased risk of heart disease, stroke, and dementia[1]. As the global population ages and the caregiver gap widens, leaving fewer young people to care for more elderly adults, AI is stepping into the void.
How My AI Companion Works: Simplifying the Tech
Under the hood, My AI Companion leverages machine learning models that have been trained on massive (i.e. think “the size of all of the content on the internet”) datasets to generate natural seeming responses to user input. Specifically, we use [Large Language Models (LLMs)2] and speech-to-speech machine learning models[3] to emulate human-like text and speech responses.
My AI Companion isn’t just a voice on a speaker; it is a consistent presence that fills the silence of an empty house. It provides the cognitive stimulation of a conversation, the reassurance of a check-in, and the utility of a digital assistant, all wrapped in a persona that feels familiar and supportive.
How Does My AI Companion Understand Seniors?
The preprocessing and training work of creating these models is largely handled by our underlying tech stack, which includes APIs from leading providers such as OpenAI. On top of this, we have developed a custom layer of context engineering that allows My AI Companion to be tailored to our specific use-case. Context engineering involves structuring system prompts, conversational history, retrieval-augmented generation (RAG), and tools used to optimize AI agent performance and efficiency[4]. For instance, a senior can simply say, “I’m feeling a bit lonely today,” or “My plant is drooping, can you help me figure out how to fix it?” and My AI Companion will be able to respond appropriately.
This layer of context engineering is what sets My AI Companion apart from more generic tools, like ChatGPT or Claude. While these tools are effective at meeting many different types of use-cases, they are not specifically tailored to any particular one.
How Does My AI Companion Decide What to Respond to Seniors?
When a user creates a new account and begins talking with My AI Companion, the companion is initialized out-of-the-box with conversation guardrails, best practices, examples, and engagement tracking. Beyond this, we have a memory system and topic exploration objectives that push the companion to actively learn and leverage knowledge about the users in patterns that are based on human companionship.
All of this logic is invisible to the end-user, but creates a more natural, human-like experience. Just as a human friend remembers your favorite tea or that you always go for a walk at 10:00 AM, the AI builds a profile of the user’s routine, preferences, and history. Over time, it can move from reactive to proactive support.
What Happens When The Senior Doesn’t Pick Up the Phone?
In addition to context engineering, My AI Companion has custom tooling to support daily check-ins and escalations and to contact family members, if necessary. For instance, if the care circle feature is enabled, My AI Companion can reach out at set intervals (e.g. daily, twice daily, twice a week) and register if the user responds or not. If it knows that it is supposed to call Jack every morning at 9:00 AM, but he doesn’t pick up the phone call or respond to the text, it can gently prompt him: “Good morning, Jack! I noticed you haven’t answered my call today. Are you feeling alright?”
If there is no answer, My AI Companion will try a few more times before alerting a list of trusted contacts designated by the user, which we call the care circle. The trusted companions in the care circle can then choose to reach out to the user to make sure that they are okay. This feature was developed based on community input.
Safety and Security by Design
The primary concern for many is privacy: “Is this thing spying on me?” At My AI Companion, we take such concerns very seriously. At the device level, we leverage the built-in privacy protections for SMS and voice calling. On the server side, we observe HIPAA Security & Privacy standards.
How Do Safety Guardrails Protect the Older Adult?
Beyond simple reminders for medication, My AI Companion acts as a sophisticated health and security monitor. By analyzing speech patterns and response times, the AI can recognize concerning patterns that might be indicative of depression, suicide risk, increased loneliness, anxiety, elevated pain levels, and lower cognitive functioning. These indicators are then used by the companion to provide more context-appropriate support. Safety is the cornerstone of My AI Companion.
What about My AI Companions and Mental Health?
In the case of suicidality, we have additional guardrails (i.e. multiple agent-as-a-judge[5] calls) to ensure that we do not miss signs of suicidal ideation. Response to suicidal ideation follows the best practices outlined by OpenAI (e.g. users are forwarded to 988 for more appropriate human support). Similar guardrails are in place for highly inappropriate or illegal behaviors, as outlined by OpenAI’s moderation guidelines[6].
Can My AI Companions be Trusted by Seniors?
In addition to providing useful feedback, safety guardrails are in place on My AI Companion to limit harmful hallucinations and other types of dangerous conversations. We are all hearing a frightening number of news stories about AI hallucinating or providing dangerous advice. To prevent this, My AI Companion is built with hard-coded ethical guardrails based on Anthropic’s published guidelines and the literature on peer-to-peer senior support networks. My AI Companion is restricted by a core set of safety protocols that prevent it from suggesting any harmful, illegal, or nonsensical actions.
Unlike unregulated bots, My AI Companion operates within a walled garden of information. It cannot be coerced into radical or unsafe behavior. Its objective function is strictly defined: to promote the health, happiness, and safety of the user. It is programmed to recognize its own limitations; if a user asks for complex medical advice, the AI is trained to say, “I am your companion, not your doctor. Let’s schedule a call with Dr. Smith to discuss that.”
Built by Researchers, Guided by Real Users
Dr. Karen Fortuna, Dartmouth assistant professor and our chief science officer, and I are committed to this work because we believe that the future will involve technology empowering people to stay in their homes, maintain their connections to others, and, through that, hold on to the wit and heart that define their personalities. This empowered aging is only achieved by building products with the people who are most likely to use them and incorporating the elements that make them feel safe. My AI Companion stands out by including users in our development process and by prioritizing safety and privacy as much as companionship.
Our goal is to create a world where no one ever has to sacrifice personal safety to avoid growing old alone.
