Designing the Care Profile

If an AI companion is ever going to support a vulnerable person, the first serious object is not the body.

It is not the voice.
It is not the face.
It is not the robot shell.
It is not even the chat interface.

The first serious object is the Care Profile.

A Care Profile is the structured, human-approved map of what the companion needs to know in order to support a person safely, respectfully, and consistently.

Not everything.

Only what matters for care.

That distinction is important.

A Care Profile is not a surveillance archive.
It is not a diary of every private moment.
It is not a behavioral database for product improvement.
It is not a file that lets the AI decide who the person is.

It is a governed care map.

Its purpose is simple:

Help the care circle remember what keeps this person safe, understood, regulated, and respected.

Why the Care Profile Comes First

Assistive technology is already meant to support functioning, independence, inclusion, participation, and dignity. The World Health Organization describes assistive products as tools that help maintain or improve functioning and independence, including support for cognition, communication, mobility, self-care, hearing, vision, and more. WHO also notes that many people need more than one assistive product, making integrated support important. (World Health Organization)

That integrated part matters.

A vulnerable person’s support needs are rarely one-dimensional.

A non-speaking autistic child may need communication support, sensory support, routine support, safety support, caregiver coordination, and privacy protection at the same time.

If an AI companion only knows one piece, it can misunderstand the whole person.

That is why the Care Profile must be layered.

Layer 1: Identity and Dignity

The first layer is not diagnostic.

It is personal.

The Care Profile should begin with the person’s name, preferred forms of address, household language, cultural context, religious or family boundaries where relevant, and dignity rules.

This layer answers:

How should this person be addressed?
What language should never be used?
What assumptions should the AI avoid?
What makes the person feel respected?
What should caregivers remember before speaking about them?

For a child, this also means the AI should not speak as if the child is absent when the child is present.

A non-speaking child is still a person in the room.

The Care Profile should protect that truth.

Layer 2: Guardian and Authority Information

The Care Profile must define who has authority.

Who is the parent or legal guardian?
Who are approved caregivers?
Who can view care notes?
Who can edit routines?
Who can approve changes?
Who can export records?
Who should be contacted in an emergency?
Who is allowed to receive summaries?

This is where Ahd Nucleus logic enters directly.

Authority must not be vague.

The AI should know the difference between:

a parent-approved care rule
a teacher’s observation
a therapist’s note
a caregiver guess
an AI-generated pattern
a raw event log

These are not equal.

A Care Profile must have an authority ladder, because in care, a confident machine guess can become dangerous if treated like truth.

Layer 3: Communication Map

For a non-speaking child, communication cannot be reduced to speech.

AAC, or augmentative and alternative communication, includes the many ways someone communicates besides talking. ASHA describes AAC as including gestures, facial expressions, writing, drawing, spelling words by pointing to letters, pointing to pictures or words, and using devices such as tablets, apps, computers, or speech-generating tools.

So the Care Profile should include a Communication Map.

It may record:

How the person says yes
How the person says no
How they ask for help
How they show pain
How they show hunger
How they show tiredness
How they show overload
How they seek comfort
How they refuse
How they use AAC, gestures, pictures, objects, sounds, or routines
Which signals are confirmed
Which signals are uncertain
Which signals require human checking

This layer should be humble.

It should not say:

“He always means this.”

It should say:

“This signal has often meant this, but human confirmation may be needed.”

Care requires humility.

Layer 4: Sensory Map

The next layer is the Sensory Map.

This is where the Care Profile records the person’s sensory world.

Which sounds hurt?
Which lights overwhelm?
Which textures are avoided?
Which textures soothe?
Which foods are safe?
Which smells are distressing?
Which environments are difficult?
Which objects regulate?
Which forms of touch are welcome?
Which forms of touch are not welcome?
Which sensory interests are safe?
Which sensory interests require supervision?

Autistic people may process sensory input differently, and sensory overload can show up through withdrawal, distress, stimming, hiding, running away, crying, shouting, or resisting touch. The National Autistic Society describes sensory overload as taking in more sensory information than a person can process.

A good Care Profile does not treat sensory needs as side notes.

It treats them as care infrastructure.

Because if the sensory map is wrong, the companion may escalate distress while thinking it is helping.

Layer 5: Routine and Transition Map

Many support needs appear during transitions.

The Care Profile should record:

Morning routine
Meal routine
Bath routine
Sleep routine
School or therapy routine
Travel routine
Prayer or family routine, if relevant
Preferred transition warnings
Visual schedule needs
First/then language
Countdown preferences
Objects that help transitions
What makes transitions worse

The goal is not to make the person obey faster.

The goal is to reduce avoidable distress.

A bad system says:

“Stop now. Do this.”

A better system says:

“Five minutes. Then bath. First tablet away, then towel.”

An even better system knows whether that script actually works for this person.

Layer 6: Regulation and Distress Plan

A Care Profile must include what helps when the person is dysregulated.

This may include:

reduce speech
lower sound
dim lights
create space
offer AAC
offer water or snack
offer a known regulating object
avoid touch
play familiar audio
move to a quieter room
call a specific caregiver
stop demands temporarily

It should also include what makes distress worse.

Too much talking.
Sudden touch.
Crowding.
Repeated demands.
Taking an object without transition.
Bright light.
Loud environment.
Unfamiliar people.
Changing the plan without warning.

This layer protects the person from being treated as “noncompliant” when they are actually overwhelmed.

Layer 7: Safety Map

The Safety Map is not optional.

For some autistic children and people with developmental disabilities, wandering or elopement can be a serious risk. The CDC describes wandering as leaving a safe area or responsible caregiver, and notes that some children and youth with autism or intellectual disability may have difficulty understanding safety issues or communicating with others. The CDC also reports that about half of children and youth with ASD were reported to wander in a parent survey. (CDC)

A Care Profile should record:

Wandering risk
Water risk
Traffic risk
Door/window risks
Unsafe objects
Food allergies or restrictions
Medical alerts
Seizure information, if relevant
Medication notes, if relevant
Emergency contacts
Safe places
Whether the person responds to name
Whether the person can identify themselves
When to alert a human immediately
What the AI must never handle alone

The companion must not improvise here.

Safety rules should be guardian-approved, source-traced, and reviewed regularly.

Layer 8: Care Circle and Setting Differences

A person may behave differently across settings.

Home is not school.
School is not therapy.
Therapy is not a public place.
A grandparent’s house is not a clinic.

The Care Profile should record setting-specific notes:

At home
At school
At therapy
In public
During travel
With unfamiliar caregivers
During illness
During holidays or disrupted routines

This matters because a pattern may only appear in one setting.

If the system flattens all settings into one generic profile, it may misunderstand the person.

Layer 9: Source Trace

Every care note should have a source.

Who observed this?
When?
Where?
How often?
Was it confirmed?
Was it reviewed by the guardian?
Is it still current?
Has it changed?

A parent observation may be deeply valuable.
A therapist note may be clinically useful.
A teacher note may reveal school-specific patterns.
An AI-generated observation may be useful but still unconfirmed.

The Care Profile should not let these collapse into one undifferentiated memory.

Source trace protects the person from stale, wrong, or overconfident care assumptions.

Layer 10: Review and Expiry

Care changes.

Children grow.
Needs shift.
Communication expands.
Triggers fade.
New risks appear.
Old supports stop working.

So the Care Profile needs review cycles.

Some notes may be stable.

Some should expire.

Some should be marked:

current
needs review
superseded
uncertain
emergency only
guardian-approved
clinician-reviewed

UNICEF’s 2025 guidance on AI and children emphasizes child-centred AI requirements including safety, protection of children’s data and privacy, transparency, explainability, accountability, inclusion, and child well-being. It also notes newer concerns around AI companions used by children and opportunities around accessibility for children with disabilities. (UNICEF)

That means a Care Profile should not be a dead file.

It should be a governed living object.

What the Care Profile Must Not Become

A Care Profile must not become:

a surveillance archive
a compliance tracker
a corporate training dataset
a behavioral scoring system
a substitute parent
a substitute clinician
a permanent label that traps the person
a place where every private moment is stored forever

The goal is not to capture the whole life.

The goal is to preserve what care genuinely needs.

This is the rule:

The Care Profile serves the person. It does not consume them.

A Simple Care Profile v0.1

A first version could be structured like this:

Care Profile v0.1

1. Identity & Dignity
- Name:
- Preferred address:
- Languages:
- Cultural/family notes:
- Dignity rules:
- Words/tones to avoid:

2. Guardian & Authority
- Legal guardian:
- Approved caregivers:
- Emergency contacts:
- Who can view:
- Who can edit:
- Who can approve changes:

3. Communication Map
- Yes signals:
- No signals:
- Help signals:
- Pain signals:
- Hunger/thirst signals:
- Overload signals:
- AAC tools:
- Uncertain signals:

4. Sensory Map
- Sound sensitivities:
- Light sensitivities:
- Texture sensitivities:
- Food/texture notes:
- Calming inputs:
- Sensory interests:
- Supervision risks:

5. Routine & Transitions
- Morning:
- Meals:
- Bath:
- Sleep:
- School/therapy:
- Preferred transition supports:
- Known transition risks:

6. Regulation Plan
- What helps:
- What makes it worse:
- Reduce demands by:
- When to call caregiver:

7. Safety Map
- Wandering risk:
- Water risk:
- Traffic risk:
- Medical alerts:
- Allergies:
- Unsafe objects:
- Emergency rules:

8. Setting Notes
- Home:
- School:
- Therapy:
- Public places:
- Travel:

9. Source Trace
- Note source:
- Date:
- Setting:
- Confidence:
- Reviewed by:
- Review date:

10. Review Status
- Current:
- Needs review:
- Superseded:
- Guardian-approved:
- Clinician-reviewed:

Where Ahd Nucleus Fits

In Ahd Nucleus language, the Care Profile is a specialized continuity object.

It combines:

Spine
Memory Vault
Authority Ladder
Source Trace
Canon Distillation
Guardian Gate
Dignity Guard

But in Amanah Companions, the stakes change.

The object is not preserving a writing project.

It is preserving care knowledge around a vulnerable person.

That means the rules must be stricter.

More privacy.
More review.
More source trace.
More humility.
More human authority.
Less assumption.
Less collection.
Less automation.

Closing

A care companion cannot begin with personality.

It cannot begin with charm.

It cannot begin with embodiment.

It must begin with the Care Profile.

Because care is not generic.

A child’s distress is not generic.
A child’s communication is not generic.
A child’s sensory world is not generic.
A child’s safety needs are not generic.
A family’s routines are not generic.

If the AI does not know the person’s care map, it should not pretend to know the person.

The Care Profile is not the whole future.

But it is the first door.

And if that door is built badly, everything after it becomes dangerous.