%20scale(.8)'%3e%3cpath%20d='M116.770%2049.147A48%2048%200%201%200%20116.770%20110.853'%20fill='none'%20stroke='%2316263D'%20stroke-width='27.5'%20stroke-linecap='round'/%3e%3cpath%20d='M138.5%2080%20A10.5%2010.5%200%201%200%20117.5%2080%20A10.5%2010.5%200%201%200%20138.5%2080%20Z'%20fill='%23F06D5E'/%3e%3c/g%3e%3cg%20fill='%233D8B83'%20stroke='%233D8B83'%20stroke-linecap='round'%20stroke-linejoin='round'%3e%3cpath%20d='M445.5%20132%20A7.5%207.5%200%201%200%20430.5%20132%20A7.5%207.5%200%201%200%20445.5%20132%20Z'%20stroke='none'/%3e%3cpath%20d='M511.5%20110A25.5%2025.5%200%201%201%20486%2084.5A25.5%2025.5%200%200%201%20511.5%20110V135.5'%20fill='none'%20stroke-width='14'/%3e%3cpath%20d='M552%2091V135.5'%20fill='none'%20stroke-width='14'/%3e%3cpath%20d='M559.5%2068%20A7.5%207.5%200%201%200%20544.5%2068%20A7.5%207.5%200%201%200%20559.5%2068%20Z'%20stroke='none'/%3e%3c/g%3e%3c/svg%3e)
20. Principal Risks and Responses
| Risk | Impact | Response |
|---|---|---|
| Wrong booking or order | Cost and schedule harm | Outcome verification, final confirmation, cancellation and compensation process |
| Changed website | Execution failure | API first, change detection, isolated retry |
| Prompt injection | Leakage or wrong action | Separate content, restrict tools and authority |
| Node compromise | Privacy breach | Device authentication, isolation, short authority, remote revocation |
| Kiosk residual data | Exposure to next user | Reset, local encryption, automatic deletion |
| Voice or wearable error | Wrong judgment | Confirmation, multiple signals, no diagnosis |
| Check-in false alarm | Unnecessary report and anxiety | Staged retry, chosen contact order, measurement |
| AI replaces relationships | Deeper isolation | Prefer real contact, usage and stop controls, field evidence |
| Robot collision | Physical or property damage | Limits, physical stop, standards, insurance, staged trial |
| Household surveillance | Privacy loss | Default non-collection, local processing, channel consent, deletion |
| Excess guardian control | Loss of self-determination | Granular authority, review records, withdrawal |
| Data-reward pressure | Exploitation | Separate consent, fair reward, minimum collection |
| Reward fraud | Higher cost | Outcome proof, anomaly detection, dispute review |
| Platform policy change | Reduced mobile capability | Standards, phone, watch, kiosk, and secure-execution alternatives |
| Uneven helper quality | Safety and trust loss | Role checks, training, evaluation, qualification limits |
20.1 When automation must stop
Stop when intent is unclear and cost or legal effect may occur; unrequested data transfer is required; a site asks for greater authority or outside contact; results cannot be verified and duplication is possible; medical, legal, or financial judgment is required; the user expresses anxiety, confusion, or refusal; a companion lacks grounds to interpret non-response; or robot sensing, braking, or connectivity falls outside safety limits.