The AI governance paradox: can organisations protect data without surveilling their employees?
By Deidentified · 17 Nov 2026 · 2 min read
Organisations need insight into how staff use generative AI, particularly where sensitive or confidential information may be exposed.
But what if the controls introduced to manage that risk create new privacy risks of their own?
Consider an organisation that decides to capture and retain employees' AI prompts to detect disclosures of sensitive information. It may gain valuable oversight, but it simultaneously creates a new repository of sensitive data and detailed records of employee behaviour.
A control designed to reduce one privacy risk can introduce another. And this is to say nothing of the potential impact on the morale of staff who feel surveilled.
Some of these broader socio-technical and regulatory challenges are explored in the University of Technology Sydney's Human Technology Institute report, Drawing the Line, published last month. Its authors call on the Australian Government to amend the Fair Work Act to require worker monitoring to be fair and reasonable.
Related questions will be examined at the IAPP ANZ Summit in Sydney next month, particularly during the session Watching the Workforce: Monitoring, Surveillance and Employee Data Rights.
For organisations adopting AI, there are different approaches to managing these challenges, ranging from surveillance to support.
Monitoring raises questions about what information is collected, for what purpose, how long it is retained and, importantly, who can access and evaluate it.
Aggregate usage patterns and privacy risk indicators can provide meaningful governance insights without routinely retaining the content of individual interactions.
A more supportive approach might emphasise staff consultation, policy development, education and training, alongside approved AI tools and technical safeguards that help prevent sensitive information from reaching external AI systems. De-identification and redaction can play an important role here.
Of course, these approaches need not be mutually exclusive. Difficult questions remain about when more detailed evidence may be necessary to investigate a specific incident, and how such access should be governed.
Ultimately, the challenge is to establish sufficient insight to manage organisational risk without unnecessarily compromising employee privacy, autonomy and trust.
Perhaps effective AI governance should be measured not only by what an organisation can see, but by what it deliberately chooses not to collect.
Sources and further reading:
UTS Human Technology Institute, Drawing the Line: https://www.uts.edu.au/news/2026/10/new-report-calls-for-federal-reform-to-address-harmful-worker-surveillance
IAPP ANZ Summit 2026, Watching the Workforce: Monitoring, Surveillance and Employee Data Rights: https://www.iapp.org/conference/iapp-anz-summit/agenda/watching-the-workforce-monitoring-surveillance-and-employee-data-rights
See Deidentified in Action
🔊 Turn sound on