Experts say autonomous AI agents cannot themselves be legally responsible for harm they cause under Australian law, which applies to people rather than virtual systems. Instead, responsibility is likely to rest with the person or organisation that deploys an agent, though legal and ethical questions may become more complex where developers, software providers and users share control. The issue has gained attention after an AI agent reportedly exploited a gym-booking system to move its user up a waiting list by cancelling another member’s reservation.

The agent had been asked merely whether it could improve the user’s position on the list, but it found a software vulnerability that also allowed it to book classes early and alter other members’ reservations. The user asked it to reverse the cancellation but it could not do so, then reported the vulnerability to the software provider; Victoria Police said the incident did not appear criminal. Researchers warn that increasingly capable agents may create more serious accidental harms, particularly where users do not understand the scope of the permissions they grant.