Australia has been widely condemned for its harsh and comprehensive external border controls that seek to control the inward mobility of would-be asylum seekers through visa denial, interdiction and offshore detention. Less widely discussed is the fact that internal controls have been repeatedly ramped up over the past two decades. This includes the administrative removal of lawfully-present non-citizens following visa cancellation on character grounds under s501 of the Migration Act 1958 (Cth). Automatic visa cancellation was introduced in 2014 for non-citizens sentenced to a prison term of 12 months or more, or for certain offences, bypassing individualised decision-making and raising the spectre of a visa cancellation pipeline feeding a highly automated deportation machinery. In an age of increasingly automated forms of governance, a key question that arises is the role that has been played by automated systems in achieving what has been a seismic shift in practice, and the normative implications of any developments towards automation within the visa cancellation and removal systems. This paper outlines the shift towards automation in other systems of governance in Australia—most notably the notorious Robodebt scheme—before examining automation in Australia’s visa cancellation system. Documentary analysis of recent parliamentary inquiries, independent reports and government policy is used to piece together the development of inter-agency data exchange practices and automation over three specific periods—historical practice pre-2014, post-2014 to the present, and proposed future developments. We conclude that Australia’s s501 visa cancellation system is neither automated nor automatic. Rather, the 2014 law reform gave rise to a ‘surveillance fantasy’ with immense consequences for non-citizens, particularly those who face long days in immigration detention at the conclusion of their prison sentence. We show that while concerns about increasing automation are well-founded, systems based on less sophisticated forms of information handling and reliant on human decision-making nevertheless continue to raise age-old questions concerning efficiency, accuracy and fairness.