Josh.ai publishes a code of conduct for AI in the home
Twelve rules and a permission model built so the owner of the house stays in charge of what happens inside it.
Josh.ai has published a Code of Conduct for AI in the Home, a first edition dated September 2026. It contains twelve rules and a permission architecture the company calls the Home Authority Stack.
Artificial intelligence in a browser writes, searches, and advises. Artificial intelligence in a house reaches the physical world. It changes the lights and the climate, opens or secures a door, watches a camera, runs the water. The document opens on that difference: a wrong answer is one thing, and a wrong physical action is another.
Josh.ai puts the document forward as a voluntary code for any AI that can observe, reason about, recommend, automate, or control connected systems in a home. The company says the bar is deliberately set higher than any product on the market meets today, its own included, because the purpose is to define where the industry needs to go.
"The intelligence in a home must always answer to the people who live there," said Alex Capecelatro, chief executive officer of Josh.ai, announcing the code.
"Home AI should be powerful, but never sovereign."
A Code of Conduct for AI in the Home, Josh.ai
Why general AI principles do not cover a house
Four things separate a home from any other place AI operates.
Actions are physical
A request can move a motor, unlock a door, or shut off water. Safety cannot stop at a good answer. It has to carry through to who was allowed to do the thing, what actually happened, and whether it can be undone.
The data is intimate
A house reveals who is present, when people sleep, which rooms they use, how routines change. The AI can infer more than any single sensor records, so privacy belongs in the architecture rather than in a policy document.
Authority is shared
Most AI is built around one account holder. A house holds partners, children, guests, caregivers, and installers. One person's request affects everyone else, so a single login is not enough.
It has to keep working
Lights, climate, and access should not become unusable because the cloud is unreachable. Core functions degrade gracefully, and local control keeps working whenever it can.
The AI does not control the house
This is the center of the document. The home AI should not be the control system. It should be one intelligent component inside a control system with fixed boundaries, and a separate policy and permission engine decides whether a proposed action is allowed before any device command goes out. Every physical action passes through that layer first.
Human controls and household rules
Permissions, roles, standing authorizations, and the stop and disable controls. Set by the people who live there.
The home AI reasons
Voice, language models, perception, planning. It works out what you want and it can suggest, but it never operates a device itself.
The control system decides
It checks the request against the permissions, the risk, the scope, and the safe states. Every physical action passes through here.
The home acts
Only what was allowed gets carried out, locally wherever it can be, and the system records what happened.
Privacy gateway, held separate from the AI
Whatever leaves the house is minimized and de-identified first, so outside services learn as little as possible about the household.
How permission flows: the Home Authority Stack
Six layers, each bounded by the one above it. The AI sits at the bottom with the least authority of anything in the house. It can narrow what it does. It cannot widen it.
Safety and security Layer 1
Constraints nobody overrides. Not a resident, not an installer, not the model, not a third party.
Household rules Layer 2
Set by whoever administers the house. Who reaches what, from where, and under which conditions.
Roles and areas Layer 3
Resident, child, guest, caregiver, installer, technician, and the rooms and devices each one may touch.
Standing authorizations Layer 4
Scenes, schedules, automations, and any proactive behavior, scoped narrowly.
Immediate request Layer 5
The instruction somebody is giving right now.
AI inference Layer 6
The AI fills in details, and only inside the authority the five layers above already granted.
What changes in practice: different people hold different permissions. Through an app the house knows exactly who is acting. By voice it applies the safest permissions that fit and asks when identity matters. An automation is not the AI deciding something, it is a person's standing instruction running inside limits they set in advance. A request that conflicts with the household rules is declined or narrowed rather than executed because the model found a clever way around it.
The twelve rules
Josh.ai frames these as proposed industry principles. Implementation can vary by product, but the outcome is supposed to be recognizable to the people living with the system. Every rule answers the same question: who is in charge here.
Humans remain in control
Authorized people can interrupt, redirect, correct, pause, or end what the AI is doing. It never resists intervention or makes stopping harder, and ongoing work does not restart on its own.
The AI acts only with authority
Every physical action traces back to a current request or a standing authorization a person created. The AI does not invent goals or use permissions it was never given.
Consequence sets the bar for confirmation
The more irreversible the action, the more certain the AI has to be and the more likely it is to ask first. Access, security, fire, and water carry a higher bar.
Dimming a light can be instant. Disarming the alarm for an unknown visitor should not be.
Off means off
You can disable the AI completely, and the path to do it is obvious. It never requires a support call. Manual control of lights, climate, and media keeps working after you switch it off.
Behavior is predictable and bounded
The AI never operates a device directly. Every request it makes goes through the control system, which checks it against the permissions in force.
The AI may decide what "make it cozy" means. The control system decides which rooms it may change.
Actions are explainable and auditable
Consequential actions leave a readable record of the trigger, the devices affected, the result, and any errors. Explanations come in plain language rather than log output.
Actions are reversible or safely contained
Undo restores the prior known state rather than firing a generic opposite command. The system distinguishes completed, pending, failed, and uncertain actions.
Privacy is the default
The AI works with the least data it can. Household data is never sold, used for advertising, or shared outside the house without permission. Cloud recording is optional, with clear deletion.
Security is the default
Security is built in rather than left to the owner or the integrator as optional hardening. No universal default passwords. Multi-factor authentication for administrative and remote access, and encryption in transit and at rest.
Local first, resilient by design
Core control prefers local paths, and locally controllable devices are preferred where practical. If the cloud is unreachable, you can still run lights, scenes, and physical controls.
Access is limited, and nothing grants itself more
Every app, device, service, and AI gets only the access it needs for only as long as it needs it. Installer and service access is explicit, auditable, and time limited. The AI cannot change its own permissions or disable its safeguards.
Uncertainty reduces autonomy
When the AI is unsure about intent, identity, device state, or consequence, it does less rather than more, and it reports uncertainty instead of pretending the command worked.
If "open it" could mean a shade or an exterior door, it asks.
Local where it matters, cloud where it helps
The document is direct about where processing belongs. Inside the house it is fast, private, always available, and it keeps working offline. In the cloud there are large models, broad knowledge, and heavy reasoning. The house decides which is which: personal and time-critical work stays local, and only what is needed goes out.
Capability can grow. Authority does not have to grow with it.
Part five takes on the worry that AI eventually improves faster than anyone can track. Josh.ai says it takes the concern seriously and has not tried to predict whether it happens. Instead the company designed the house so that it would not change who is in control. However capable the model becomes, it still answers to the same control system, the same permission layers, and the same off switch, which the document says should stay physical and absolute.
There is a second half to that argument, and it is the more interesting one. The safety rules themselves should be held apart from the model, so that neither an intruder nor the system can weaken them, switch them off, or rewrite the limits that keep it accountable. An AI able to edit its own limits is not contained, however well it behaves.
Josh.ai also says industry cannot carry this alone, and that AI in the home deserves the same seriousness given to building codes and electrical standards. The company is offering the code as a starting framework and says it is willing to work with policymakers, researchers, and competitors on what comes next.
Where the thinking came from
The code lists the work behind it. It was not written from scratch. Asimov's laws of robotics get credit as the founding intuition that capability should be subordinate to human command, alongside Ann Cavoukian's Privacy by Design, the Asilomar AI Principles, the OECD and UNESCO frameworks, the NIST AI Risk Management Framework, device security baselines including NIST IR 8259, ETSI EN 303 645 and CISA Secure by Design, the EU AI Act, OpenAI's Model Spec, and Microsoft AI's Humanist AI Code of Conduct.
The part about integrators
One section of the code is aimed squarely at our end of the business. A modern house is not a single device, it is an ecosystem of lights, climate, locks, cameras, audio, networks, and now AI, all expected to work together safely. Josh.ai argues that getting that ecosystem to actually meet a standard like this one is real work: secure configuration, sensible roles and permissions, no default passwords, local-first device choices, and a graceful fallback when something goes offline.
Anyone can set up a smart home, and plenty of people do it themselves. The code takes the position that wherever it is practical, a certified professional integrator in the mix makes the house meaningfully safer and more reliable, and it points to industry bodies such as CEDIA that train and certify integrators to build in security, privacy, and resilience from the start. Interseckt has been a CEDIA member for a long time and holds Home Technology Association Estate Tier certification.
The code also makes a recommendation we would like to see adopted widely. Because keeping a house secure is ongoing work, a homeowner should be able to delegate limited administrative access to a trusted certified integrator, including the ability to approve multi-factor or remote access. That delegation is granted for a set time, scoped to what is needed, and fully auditable, so expert help never turns into permanent control of somebody's house.
What this means for the systems we design
Permissions are not something you bolt on at the end of a job. Which door a guest account can unlock, or what a contractor reaches during a two week remodel and loses on the last day, belongs in the design phase next to the rack elevation and the network plan. It is also the reason clients call us later: a new resident to add, a guest's access to revoke, a retired device to remove, a permission to tighten.
Josh.ai is one of the home operating systems we install, alongside Crestron, Control4, Savant, and Basalte. A platform maker putting its rules in writing helps. It gives the owner, the design team, and us something concrete to point at, and something to hold manufacturers to. That holds whichever platform ends up in the house.
This article is based on A Code of Conduct for AI in the Home, first edition, September 2026, published by and copyright Josh.ai, and on the announcement the company issued on September 23, 2026. All credit for the code, the twelve rules, and the Home Authority Stack belongs to Josh.ai. Quotations are drawn from those documents. Josh.ai notes that the code is a set of design and governance principles, not legal advice, a product warranty, or a certification that any system meets every recommendation.
Read the code in full at josh.ai/ai. The company takes feedback on it at [email protected].
Talk to us about AI in your home
If you are building or renovating and want to understand how voice, automation, and permissions should work in your house, we are glad to walk you through it. We design and install these systems across South Florida and the Caribbean.
Tell us a little about the project and we will get back to you.