Imagine buying a household robot. You want it to remember where things are in the kitchen. Your partner wants no recordings in the bedroom. Your parents want help whenever they need it. A visiting friend does not even know it is identifying voices.
There is one purchase button, but many people are affected.
This reveals a central challenge for household AI:One person paying should not mean one person decides for everyone how they are observed, remembered and acted upon.
It also makes the home one of Physical AI's most complex deployment environments. People in the same room may share a task goal but have different privacy boundaries. Close relationships do not mean every piece of information should be shared.
Buying a device involves three different decisions
The first decision concerns ownership and upkeep: who buys, repairs and renews. The second concerns information: whose voices and images are processed, and who can see the history. The third concerns relationships: whether the robot can intervene in arguments, relay conversations to others or make arrangements on someone's behalf.
Giving all three to a single administrator simplifies the interface, but may complicate life.
For example, a family member who wants a robot to track their diet may not want everyone to inspect it. Adult children buying a device for their parents does not mean the parents welcome continuous observation. Living together requires coordination, but coordination need not erase personal boundaries.
An Oxford Internet Institute overview of digital care research notes that technology can support remote care while also creating tensions around privacy, monitoring and relationships. This material cannot predict the outcomes of every household robot, but it shows why saying it is for someone's care does not automatically settle questions of permission.Oxford research overview
From a single account to a shared space
Much software is organized around accounts: whoever logs in is treated as the current user. A household robot's sensors, however, face a space. That space may contain family members, visitors, repair technicians and care workers. They may never have registered an account or read the terms.
I therefore suggest designing household permissions around four questions: who can give commands, who can see results, who can delete records, and who can pause sensing or movement?
These permissions need not be identical. Allowing visitors to request that filming stop need not let them disable every safety function. Allowing family members to request that objects be moved need not give them access to private conversations. Products should distinguish permissions for actions, information and maintenance.
Local operation and external access need particular distinction. Once remote support staff can assist a robot, households need to know when the connection begins, what information those staff can access and who approves it. The expert-assistance mode described for 1X's NEO shows that this is already a practical issue in public product designs. It does not establish that other manufacturers have the same arrangements.1X official product information
People can only understand what they are authorizing if they can see who is currently involved.
Household memory needs more than one drawer
A robot may need to remember where dishes belong to complete a task. To offer personal help, it may also remember someone's preferences. Those two kinds of memory should not default to the same sharing rules.
Household memory could have three layers: information needed for shared tasks, information individuals choose to save, and information processed briefly for the task at hand. A shared shopping list noting low milk supplies and someone's late-night expression of distress do not have the same need for retention.
This is a product-design proposal, rather than an existing industry standard. Its aim is to reduce the burden of granting permission: clear defaults can require more explicit decisions for highly sensitive actions without asking people to read complicated settings every day.
For children, people with limited cognitive capacity and other household members who need support, arrangements require particular care around understanding, caregiving responsibilities and specific legal requirements. A checkbox in an ordinary account cannot carry all those judgments.
Household changes need corresponding procedures. When someone moves out, a caregiver changes, partners separate or the device is sold, whose history remains in the machine? Permissions set once on installation day cannot keep pace with real life.
The strongest objection: complex permissions make products harder to use
The objection is valid. If every action needs everyone's approval, a robot may become impractical. In care settings, it might even fail to respond in time.
The answer should be permission proportionate to risk. Routine cleaning agreed in advance can proceed automatically within clear limits. Private rooms, remote viewing, sensitive records and physical contact need stricter rules. Emergency responses should also have pre-agreed triggers, notification recipients and records that can be reviewed afterward.
We cannot ask AI to understand complex human lives while designing permissions as though every household will always have a single shared wish.
Household members can also have a brief discussion before buying: which spaces are off limits, what should not be saved, who can connect, and what should pause first when people disagree? The rules need not become a lengthy contract, but everyone should be able to understand and revise them.
AI will change households; our choices shape the direction
Theoretical research on intimate AI reminds us that personality continuity, emotional responses and commercial structures jointly shape the experience. It offers a framework for discussing new relationships in the home. It does not directly prove that AI will increase divorce, reduce fertility or replace a particular family structure.Intimacy by Design
More testable changes include whether family members argue less about housework or more about monitoring; whether care becomes easier or someone must remain constantly online; and whether technology respects members who do not want to use it.
Founders can include these questions in product testing. Ask not only the purchaser, but also other residents and caregivers about their experience. For consumers, convenience worth buying should be able to coexist with the different boundaries of people in the home.
Product teams can test defaults through concrete situations. How does the system respond when an unregistered visitor asks it to pause? What limits does a household administrator encounter when seeking another adult's history? If the device cannot identify the speaker, does it proceed or ask for confirmation? These tests examine how power is distributed as well as recognition accuracy. Household members should be able to understand recent sensitive actions without receiving full transcripts of other people's conversations. Transparency should support accountability without creating a new instrument of household surveillance.
The most mature household AI may be one that knows when to ask, when to forget and when it must stop, rather than one that knows everything happening at home.
If you buy an AI, should other household members have the right to say no to it? Which rights should exist by default?