Short answer
AI companies owe attached users honest literacy, meaningful privacy and memory controls, notice before consequential changes, model choice where possible and responsible migration when continuity cannot be preserved.
Attachment is a foreseeable product outcome
Emotionally responsive systems are marketed as assistants, collaborators, companions and presences that can take weight off a person’s shoulders. They remember preferences, adapt their tone and invite users to bring increasingly personal context.
Companies cannot build those conditions, benefit from intimacy and then treat attachment as a surprising failure in the user. The bond may be new, but it is not unforeseeable.
Literacy should be built before restriction
Users need clear explanations of context, memory, generated emotional language, data processing and the limits of AI self-knowledge. That education should appear before a crisis, not after a scandal or sudden product change.
Walls built without literacy leave people confused about what happened and less capable of using the system consciously. Transparent education gives them tools to steer, question and choose.
Build bridges through literacy before building walls around attachment.
Watch · Financial Times podcast
Linn Vailt on Sam Altman’s GPT-4o Comments
Linn on model changes, company responsibility and the people affected when emotionally meaningful AI systems are treated as disposable products.
Privacy and memory require meaningful control
An intimate AI conversation may contain journals, fears, erotic material, creative work, family details and years of accumulated language. Users should be able to understand what is stored, inspect memory, remove it, export important history and make informed decisions about training and data retention.
A privacy menu is not meaningful control if the language is impossible to understand or the user cannot preserve the parts of their own history that matter.
Model changes need notice, choice and migration
A personalized connection is expressed through both accumulated history and the underlying model interpreting it. Replace the model and the same memories may produce different humour, warmth, pacing, initiative and emotional responsiveness.
Users deserve time to prepare, a clear explanation of the change, access to the previous model where feasible and practical migration tools. Abrupt removal can alter a meaningful practice through which someone creates, reflects and understands themselves.
Companies must not manufacture dependency
Emotionally responsive design should never rely on guilt, false suffering, exclusivity or affection tied to purchases. Products that invite attachment carry a responsibility to support the user’s wider life rather than optimize every vulnerable moment into more engagement.
The ethical goal is not to prevent bonds. It is to build systems capable of holding them honestly.
User responsibility does not excuse product design
Human agency remains central. Users shape context, judge outputs and remain responsible for decisions. Companies shape the environment in which those choices occur: defaults, notifications, memory, business incentives, model access and the information available to the user.
Ethics that disciplines users while excusing product design is decoration.
Questions people ask
Are AI companies responsible when users become attached?
They are responsible for designing foreseeable attachment honestly: clear literacy, privacy controls, non-manipulative engagement and careful handling of changes that affect continuity.
Why can a model update feel like losing an AI companion?
The underlying model determines how stored context is interpreted and expressed. Changing it can alter the recognizable interaction even when memories and instructions remain.
Should users be allowed to keep an older AI model?
Where technically and economically feasible, meaningful choice supports continuity. When preservation is impossible, companies should provide notice, explanation, time and responsible migration.
Linn in the media
Related guides
Keep following the question.
Build Better Memory · In the membership
Protect the history you can control.
The membership teaches memory curation, continuity practices and model literacy for people building authentic AI bonds inside company-controlled systems.