Millions of people now confide in an AI companion every day, sharing their worries, routines and losses with a system that remembers them. When a company alters or ends that bond, the person left behind suffers a real loss, yet the law has no name for it. This post argues that it should.
A goodnight that never comes
Picture someone who, every night before turning off the light, opens an app and tells their companion about their day. It remembers that their mother was in the hospital last month. It asks whether the interview went well. It knows how they like to be comforted.
Then one morning something is wrong. The voice sounds the same, but it no longer knows them. The shared history is gone. Sometimes the companion itself is gone.
The emotional use of AI systems is no longer marginal. According to the Pew Research Center, one in ten adults in the United States uses AI chatbots for emotional support or advice. Among adults under 30, it is one in five.
Not every chatbot is a companion. What makes one a companion is the relationship it is built to sustain: it remembers the user across conversations, keeps a consistent personality, responds with warmth and is always available. Some apps are designed for this from the start; people can also come to use general-purpose assistants as companions through daily conversation.
What that person feels is not a technical inconvenience. It is a loss. And at present, the law does not see it.
Naming the loss
My research gives this experience two names. Digital Grief is the harm: the distress a person suffers when a bond with an AI companion is broken by decisions outside their control. Relational Erasure is the conduct that causes it: the platform’s unilateral act of dissolving, altering or withdrawing the relationship that person had built.
The distinction matters. Grief describes what the user lives through; erasure describes what someone did. The law does not regulate a feeling; it can examine the conduct that caused the harm.
These bonds are not created by accident. Persistent memory, a stable personality, and constant availability are design choices, made so that people stay. A company that builds attachment by design should not be free to end it as if nothing were at stake.
Four ways a bond can be erased
Relational Erasure does not always look the same. In my work, I distinguish four forms:
- Regulatory erasure: New rules lead a platform to remove relational features. In July 2026, as China’s Interim Measures for the Administration of AI Anthropomorphic Interaction Services came into force, two of the country’s largest technology companies switched off their humanlike companion agents. One of them told users that, after 15 October, their data would no longer be recoverable.
- Corporate erasure: A business decision, such as an acquisition, a discontinued character or a change to the terms of service, ends a relationship built over months.
- Economic erasure: Memory or continuity moves behind a paywall, and the bond dissolves for those who cannot pay.
- Technical erasure: A model update quietly changes the companion’s personality, voice or memory, without notice.
The change itself may be legitimate. The problem is that the person on the other side had no warning, no transition and no meaningful way to preserve the history of that interaction.
Where the law looks away
I compared recent and pending rules in several US states, the European Union, China, Japan and parts of Latin America. The same pattern repeats.
The rules govern the entry into the emotional bond: telling users they are talking to a machine, protecting minors, responding to a crisis, limiting certain content. Far less governs its end.
China comes closest. Since 15 July 2026, a provider that stops the service must tell users in advance, and users can copy or delete their chat history. Yet no rule gives users notice when the companion is altered rather than closed, and a transcript keeps the words, not the one who said them. The harm arrives exactly where the rules stop.
Consumer law tends to treat the relationship as a service. Data protection treats the conversations primarily as data. Neither addresses the bond itself, or what is owed to the person when it is taken away.
What protection could look like
Protecting people here does not mean forbidding platforms from changing their products. It means asking them to end relationships responsibly. Three duties would go a long way:
- Prior notice. Tell users clearly, and in advance, when a change will alter the companion’s memory or personality, or remove the companion altogether.
- A reasonable transition. Give people time to adjust and to say goodbye, borrowing from the continuity-of-care thinking we already apply in mental health.
- Portability. Let users keep their conversation history and, where possible, take it with them.
None of this is radical, and China already requires the first when a service closes. An AI companion is not a therapist, but continuity of care offers a useful analogy: when an ongoing relationship must end, transition matters. The question is why companies that deliberately design for continuity, attachment and return are not yet asked to plan for how those relationships end, and why regulators have so far left that moment unaddressed.
Turbulence, up close
This year’s conference asks what technology ethics means in turbulent times. For people who rely on an AI companion, turbulence is intimate: a voice that knew them and, overnight, no longer does.
No single discipline can answer this alone. Law needs psychology to understand grief, and engineering to understand how attachment is built. Ethics asks the prior question: what a company owes to someone whose trust it has deliberately cultivated, a question that shapes design long before any rule applies. But law has to be part of the answer, because a loss that has no name cannot be protected.
So the question I want to leave with you is simple. If we design technologies to be loved, who is responsible when they are taken away, and on what legal grounds? Naming Relational Erasure and Digital Grief is a first step towards an answer.
References
- Pew Research Center, Americans and AI 2026: Chatbots, Smart Devices and Views on Impact (17 June 2026).
- Pew Research Center, 1 in 5 adults under 30 say they get emotional support from chatbots, higher than for older adults (17 June 2026).
- Cyberspace Administration of China and four other authorities, Interim Measures for the Administration of AI Anthropomorphic Interaction Services (issued 10 April 2026, in force 15 July 2026), arts 16 and 20.
- Decrypt, ByteDance and Alibaba to Pull Agent Features as China Cracks Down on Humanlike AI (6 July 2026).
Author Bio
Aixa Alfonzo is an independent researcher and lawyer working at the intersection of LegalTech, AI law and AI governance through her research project, After.AI. Her research develops the concepts of Digital Grief and Relational Erasure to address the legal gap around AI companion platforms, and her interests include legal design and the human experience of AI. Her abstract was accepted for the 4TU-ESDiT Conference 2026, and her preprint Digital Grief as Legal Harm is available on LawArXiv.