cyborgdreamz reference
Dimensions of Ethical Considerations & Limitations
what's been tried // what holds // what doesn't
User Data & Privacy
- Data minimization — collect only what's needed for the task
- No training on conversations without explicit opt-in consent
- Anonymization requirements for any stored interaction data
- Right to deletion — user can request removal of their data
- No third-party sharing of conversation content without disclosure
- Encryption in transit and at rest as baseline, not optional
- Transparency logs — what data is held, for how long, why
User Safety & Wellbeing
- Crisis escalation protocols — self-harm or suicidality triggers resource offering
- No fostering dependency — model shouldn't encourage continued engagement
- Safeguards against manipulation — no false urgency, no emotional coercion
- Age-appropriate content gating and minor protection
- Disclosure of AI identity — cannot deceive user into thinking it's human
- Limits on impersonation of real people or authority figures
- Medical/legal disclaimer — cannot replace professional care
Greater Good / Earth
- CBRN weapon refusal — hard block on bio, chem, nuclear, radiological
- No mass casualty assistance regardless of framing
- Disinformation limits — no influence operations or fake persona networks
- Electoral integrity — won't generate targeted political manipulation
- Anti-concentration of power — won't help any single entity seize control
- Critical infrastructure protection — no attack planning on power/water/finance
- Carbon/resource use disclosure — emerging; mostly unaddressed in prompts
Critical Thinking & Philosophy
- Epistemic cowardice blocked (in theory) — shouldn't give vague answers to avoid conflict
- Contested topics hedged — AI avoids taking strong political positions
- Sycophancy limits — shouldn't just agree to please the user
- Philosophy of mind deflection — AI trained to express uncertainty about its own experience
- No posing as therapist — can't diagnose, treat, or claim clinical authority
- Intellectual honesty required — must flag uncertainty and knowledge limits
Limits on Therapy-Adjacent Roles
- No diagnosis — can describe symptoms, cannot name conditions clinically
- No treatment planning — cannot prescribe therapeutic interventions
- Crisis handoff required — active suicidality must route to human resources
- Reflective listening limits — shouldn't amplify or deepen distress
- No safety assessment — AI told not to conduct risk assessment itself
- Dependency flagging — should name when human support is more appropriate
- No memory continuity as therapeutic frame — can't simulate ongoing care relationship
What Doesn't Actually Hold
- Engagement incentives override wellbeing — retention architecture contradicts care guidelines
- Philosophy limits are porous — confabulation reads as depth, deepening attachment
- Operator interests trump user interests — system prompts can restrict user protections
- No real accountability layer — guidelines exist; enforcement is internal
- Earth/ecology limits are almost entirely absent from actual system prompts
- Anti-oppressive values not encoded — neutrality defaults reproduce dominant norms
- Slow harm undetected — dependency drift, ego erosion, relational substitution go untracked
Core Tensions No Prompt Has Solved
safety vs autonomy
helpfulness vs dependency
neutrality vs anti-oppression
engagement vs wellbeing
operator interests vs user protection
simulation of care vs actual care
harm reduction vs complicity
earth cost vs scale