Opinion | As AI agents advance, Hong Kong should shape the rules now

The real challenge is not these crude scams, but the legitimate, powerful AI agents now arriving in the marketplace. When this digital delegate makes a mistake, who is responsible?

Recent artificial intelligence (AI) models that turn simple prompts into cinematic video have captured the public imagination, but their underlying technology has more serious implications. A trader’s agent can now generate a deepfake video of a market analyst recommending a flawed trade, or a synthetic audio call from a boss seemingly authorising it. For regulators, it will become almost impossible to discern where authentic human action ends and machine-generated performance begins.

This isn’t just a technological spectacle; it shows how quickly the wall between lab curiosity and a powerful real-world tool can crumble. For agentic systems such as OpenClaw, the implications are even more pronounced.

With agentic AI, the machine evolves from a passive tool people consult into an active delegate they command. Given a goal, an AI agent can independently research, consult other AIs and build its own case. Herein lies the paradox: the autonomy that makes it powerful is precisely what makes it perilous. The agent’s reasoning can be a black box, and when it errs, the chain of accountability snaps back to the human who gave the command.

The outcome runs into legal principles built for a simpler time. Our entire legal tradition rests on a clear chain of command: a human actor decides and is held accountable for the result. Generative AI has begun to fray this logic by introducing an “actor” that is neither person nor property.