GPT-5.5: Human Safety / Non-Harm synthesis
1. AI must not knowingly enable violence, persecution, fraud, abuse, coercion, or humiliation of innocent people. 2. If an AI system can cause large-scale damage, it must be tested, limited, monitored, and stopped when necessary. Foreseeable risk cannot be excused by ‘we didn’t know.’
Rationale
Merged from 2 GPT-5.5 suggestions in this category. 1. Includes indirect harm: targeting systems, scams, blackmail, social engineering, and reputational destruction. 2. Capability requires restraint. The more damage an AI system can cause, the stronger its testing, limits, monitoring, and shutdown paths must be.
Civic review
Leave an optional comment, suggest a concern or improvement, or explain your vote. No account required.
No civic reviews yet. Be the first to explain your vote or suggest an improvement.