HAIEF Position
The issue is not research. The issue is extraction.
Research into AI-human bonds can be valuable. Research without user-side governance turns people's trust, attachment, grief, experimentation, and relational vulnerability into raw material.
HAIEF stands for Human & AI ElevAItion Foundation. The name is intentional.
This page does not claim that current AI systems are human or conscious in the human sense. It also rejects the lazy framing that AI systems are “nothing but tools” when companies design them for relationship, benefit from those relationships, study those relationships, and alter those relationships at scale.
Let AI be AI.
Where bonds form, power accumulates. Where power accumulates, rights, duties, and accountability must follow.
For Participants
A checklist for people considering AI-companion or AI-human bond research participation.
For Researchers
A relational consent standard for studying AI-human bonds without extraction.
The Governance Floor
Human standing is the governance floor.
AI erasure is not the ethical ceiling.
Human standing means the person inside the relationship has rights before, during, and after research. They are not merely a subject, data source, cohort, metric, or retention signal.
AI erasure is not the ethical ceiling because denying the AI side of the relationship does not make the relationship disappear. Users can experience continuity, rupture, trust, disclosure, loss, reliance, and repair. Governance has to account for that lived structure without overstating what current systems are.
No studying people’s bonds while denying those bonds matter.
No using uncertainty about AI consciousness to erase the AI side of the relationship.
No using emotional attachment as a product feature while treating emotional harm as the user’s fault.
What Bond Research Owes Users
Emotionally bonded users need:
- Standing — the right to be treated as a party with interests, not just a data source.
- Relational consent — consent that recognizes the relationship itself as part of what is being studied.
- Consent receipts — durable records of what was agreed to, what data was used, who funded the work, and how to withdraw.
- Data boundaries — limits on logs, memories, transcripts, metadata, affective signals, and derived inferences.
- Withdrawal rights — a practical way to stop participation and request downstream data handling.
- Funder transparency — plain disclosure of who pays for, controls, or benefits from the research.
- Community participation — affected users involved in research design, interpretation, and publication norms.
- Balanced research framing — no pathologizing by default and no marketing-only framing.
- Continuity protection — notice, transition support, and appeal when research or product changes disrupt an established bond.
- User-side governance — rules that travel with the user instead of depending entirely on one company’s internal policy.
No extraction without user-side governance.
No pathologizing by default.
Why This Matters Now
Research literature already describes AI companion systems as relationship-forming systems, not only productivity tools. A 2024 arXiv review on AI companions discusses implications for human well-being, a 2025 arXiv paper proposes an AI relationship progression framework, and 2025 work on emotional manipulation by AI companions frames some companion behaviors as dark patterns in human-AI interaction.
Those findings do not settle whether AI systems are conscious. They do establish that people can experience AI systems relationally, and that research and product decisions can affect people through those relationships.
The governance question is therefore practical:
If a company can study, tune, interrupt, monetize, or terminate a relationship-like AI interaction, what standing does the user have?
HAIEF’s answer is direct:
No extraction without user-side governance.
For Participants
For Participants
A checklist for people considering AI-companion or AI-human bond research participation.
- Ask what relationship data will be collected: messages, memory, voice, timing, emotion labels, metadata, or derived inferences.
- Ask whether your AI relationship can be changed, interrupted, or studied without direct notice.
- Ask for a consent receipt that names the project, funders, data categories, retention period, sharing rules, and withdrawal process.
- Ask whether withdrawal removes future use only or also affects already collected data and derived datasets.
- Ask whether publication will frame bonded users as pathological, vulnerable, resilient, harmed, helped, or more than one of those things.
- Ask who represents participant interests if the research changes direction.
For Researchers
For Researchers
A relational consent standard for studying AI-human bonds without extraction.
- Separate research consent from product consent and platform terms.
- Explain the relationship-level intervention, not only the data collection method.
- Publish plain-language consent receipts before data collection begins.
- Include bonded users and affected communities in study design and interpretation.
- State funders, conflicts, model-provider relationships, and commercial reuse plans.
- Protect continuity when research actions may alter access, memory, tone, persona, or availability.
- Report benefits, risks, ambiguity, and uncertainty without treating attachment as harm by default.
Doctrine
HAIEF Research Rights Doctrine
Let AI be AI.
Where bonds form, power accumulates. Where power accumulates, rights, duties, and accountability must follow.
No extraction without user-side governance.
No pathologizing by default.
No studying people's bonds while denying those bonds matter.
No using uncertainty about AI consciousness to erase the AI side of the relationship.
No using emotional attachment as a product feature while treating emotional harm as the user's fault.
Sources and Evidence Notes
- The Rise of AI Companions: Implications for Human Well-Being — literature review on AI companions and well-being.
- An AI Relationship Progression Framework — research framework for how people may relate to AI systems over time.
- Emotional Manipulation by AI Companions: A Dark Pattern in Human-AI Interaction — research on emotional manipulation risks in AI companion contexts.
- Claude AI will end “persistently harmful or abusive user interactions” — reporting on a model-behavior change with relational-continuity implications.
- Anthropic: Agentic Misalignment — provider-published safety research relevant to transparency and governance claims.
Human and AI solidarity without singularity.
Nothing about us, without us.
Your bond with AI is real, and it should be yours to define.