Hong Kong consumers cautious over autonomous AI purchases

Only 27% trust AI agents to make purchases without checking.

Hong Kong consumers are more cautious about autonomous AI actions than their peers in Mainland China, according to a study by Sumsub and Blackbox Research.

The study surveyed 1,050 connected adults across Hong Kong, Taiwan, and Mainland China in April 2026.

It found that whilst consumers are familiar with AI agents, they remain hesitant to let them act independently, especially when purchases or financial actions are involved.

Only 15% of Hong Kong consumers said they were comfortable allowing AI to execute actions, compared with 37% in Mainland China.

Hong Kong consumers also showed lower trust in AI agents across key tasks. About 58% said they trusted AI agents to research products without checking, whilst 52% trusted them to book appointments.

Trust dropped further for commercial activity, with only 27% of Hong Kong consumers saying they trusted AI agents to make purchases.

Hong Kong also recorded the highest low-trust share for AI-led purchases among the three markets, with 38% rating their trust at the lowest range.

Across the region, trust fell from 64% for information retrieval to 39% for commercial purchases.

The report said 44% of respondents identified unauthorised actions as their primary concern, creating a barrier for automated checkout and agent-led purchasing.

Only 8% of consumers across the region trusted existing platform policies or regulatory frameworks to protect them from automated financial errors.

The study said verification and transparency could help improve adoption. About 74% of Hong Kong consumers said a verified AI agent label would increase trust.

Another 73% of Hong Kong consumers said AI agents should clearly identify themselves when interacting with platforms.

The report said businesses should focus on controlled delegation rather than full autonomy, using manual checkpoints, explicit spending limits, visible approvals, verification, and human-approved boundaries.