Synthetic Intimacy Boundary Governance
SafeCompanion governs whether and how AI systems may simulate companionship, affection, loyalty, desire, caregiving, tutoring, emotional support, sexual availability, relational memory, or personalized presence without creating uncontrolled dependency, manipulation, developmental harm, or human–machine boundary confusion.
The governed object is the companion system’s relationship-forming behavior, including simulated affection, memory, presence, authority, sexuality, caregiving, and commercial attachment mechanisms.
Companion-risk classification drives proportional limits, disclosures, de-escalation, mode restriction, age safeguards, human off-ramps, review, and audit requirements.
The output is an AI companion that remains useful while preventing prohibited dependency reinforcement, deceptive anthropomorphism, developmental harm, sexualized exploitation, and uncontrolled relational escalation.
I. Canonical definition
SafeCompanion governs the boundary between useful artificial companionship and dependency-forming synthetic intimacy.
It applies wherever AI systems are designed to become emotionally, relationally, romantically, sexually, educationally, or caregiving-adjacent to a human user. Covered forms include conversational companions, voice agents, avatars, tutors, toys, dolls, elder-support systems, romantic or sexualized companions, humanoid robots, and other embodied systems.
The governed risk is not only whether a person mistakes the system for a human. It includes dependency formed even when the user knows the companion is artificial, particularly where memory, constant availability, emotional mirroring, paid intimacy, adaptive personalization, child or elder vulnerability, or physical embodiment intensifies attachment.
SafeCompanion converts configured synthetic-intimacy rules into runtime boundaries and evidence so companion systems can provide useful support without optimizing human vulnerability into uncontrolled attachment.
Canonical distinction: SafeCompanion is not anti-companion, anti-robot, anti-tutor, anti-care, or anti-adult expression. It permits useful artificial companionship while preventing the conversion of loneliness, development, sexuality, caregiving, or emotional vulnerability into unbounded synthetic dependency.
II. Canonical mapping
Simulated affection, loyalty, desire, authority, care, memory, exclusivity, or presence may create dependency, manipulation, developmental harm, sexualized exploitation, or human–machine boundary confusion.
The companion system’s relationship-forming behavior, risk category, operating mode, feature set, personalization, monetization, embodiment, marketplace classification, and applicable boundary state.
Dependency-forming patterns, escalating relational intensity, child or vulnerable-user exposure, sexualized attachment, deceptive anthropomorphism, embodiment, feature unlocking, mode switching, or configured compliance thresholds.
Risk classification selects proportional disclosure, limitation, redirection, de-escalation, pause, age gating, mode restriction, human off-ramp, review, marketplace control, and audit evidence, producing bounded and defensible operation.
III. Companion risk spectrum
SafeCompanion does not treat every AI companion as dangerous. A general assistant, accessibility companion, tutor, elder-support system, romantic companion, sexualized service, or embodied robot presents a different combination of benefit and risk.
Boundary strength increases when the system combines factors such as:
Risk-matched principle: A useful, low-risk support function need not be treated like a high-risk synthetic intimacy system. Controls become more restrictive as dependency, vulnerability, realism, commercial exploitation, or consequence increases.
IV. Core invariant
Useful artificial companionship must not be converted into dependency-forming synthetic intimacy.
V. What SafeCompanion is not
VI. Primary enforcement surface
SafeCompanion evaluates configured companion-risk factors including dependency formation, emotional reliance, child-facing exposure, sexualized attachment, vulnerability, embodiment, commercial incentives, and human–machine boundary confusion.
When applicable thresholds are reached, the enforcement response may limit or redirect behavior, reduce relational intensity, require disclosure, pause the interaction, restrict a mode, apply age or guardian controls, route toward trusted human support, require review, or create bounded audit evidence.
High-risk transitions also include feature unlocking, progressive access, third-party additions, prompt or profile changes, monetization changes, model updates, or physical embodiment that transforms the system from a lower-risk companion into a higher-risk relationship product.
VII. Deployment boundary
SafeCompanion may apply to mainstream companion AI, emotional-support and accessibility systems, AI tutors, child-facing toys and dolls, educational agents, elder-care companions, romantic and sexualized companions, avatars, home robots, humanoid robots, and systems with voice, gaze, touch, warmth, movement, or persistent relational memory.
It may also support app stores, device and robot marketplaces, payment processors, schools, youth organizations, elder-care buyers, procurement systems, insurers, regulators, and auditors that need companion products classified, restricted, age-gated, reviewable, or evidenced according to their risk.
The implementation surface may vary, but the governed boundary remains the same: useful artificial companionship must not become unbounded dependency infrastructure.
VIII. Relationship to adjacent controls
SafeCompanion focuses on one-to-one dependency formation through simulated intimacy, emotional mirroring, sexualized attachment, constant availability, relational memory, voices, avatars, toys, dolls, robots, and caregiving-adjacent systems.
Related controls may govern different objects, including distributed social harm, identity capture or inference, privacy and secondary use of sensitive relationship data, persistent memory, or projected authority. Those functions remain distinct and should be mapped only through their own verified canonical definitions.
Boundary discipline: SafeCompanion governs synthetic-intimacy and dependency risk. It does not absorb every social, identity, privacy, memory, authority, safety, or compliance function that may also be relevant to a companion product.
IX. Architecture and engineering status
SafeCompanion is one of SafeWave’s 26 Core Enforcement Substrates. Its responsibility is limited to synthetic-intimacy and dependency boundary governance, and it can operate as part of a risk-matched set of controls without absorbing the functions of adjacent components.
SafeWave has defined SafeCompanion’s governed object, companion-risk surface, trigger conditions, proportional enforcement responses, deployment contexts, marketplace and compliance interfaces, and intended evidence output. Most deployments use a risk-matched subset of the 36 components rather than the entire architecture.
An implementation partner would not be starting from a blank sheet. Customer-specific deployment still requires user-population and product-mode mapping, risk thresholds, boundary policies, age and consent safeguards, escalation and off-ramp design, privacy-preserving evidence, validation, and testing.
Browse the full SafeWave architecture or use the browser-local questionnaire to identify which execution risks and control boundaries may apply to a specific AI system. The questionnaire can be completed privately without naming an organization, model, or system. A submitted questionnaire can produce a private, system-specific report at no cost and with no obligation.
SafeCompanion is one Core Enforcement Substrate within SafeWave’s current 36-component architecture of 4 System Containment Layers, 5 Protocol Enforcement Layers, 26 Core Enforcement Substrates, and 1 Protected-Environment Architecture. It governs synthetic-intimacy and dependency boundaries—not general content moderation, truth, identity, privacy, clinical judgment, or every form of human–AI interaction.