Here we are again, discussing the right to erasure because on 31 August 2026 the Attorney-General ruled out banning smart glasses, citing accessibility reasons.
So now our government grapples with the privacy concerns that millions of us already have with non-consensual recording and facial recognition technology, already being used at Bunnings and trialled at Woolworths and Coles.
Of course an argument (however weak) can be made that there is an element of consent when we choose to consume the services or products from these retailers.
When it comes to smart glasses, that argument fails because the women and girls in Australia, who are most at risk from this technology, are not getting any service in return. We only receive a disproportional risk to our safety and rights to our bodies.
This isn’t rhetoric, it is quantifiable in numbers. So let’s do that math.
The genuine need this technology solves is the ability to have a camera interpret scenes/text for people who cannot otherwise do this. This has two key groups, vision impaired individuals and experts who need hands free recording in dangerous or high stress environments.
From ABS data, roughly 0.6% to 3% of the Australian population is vision impaired, and taking the higher estimate that is 840,000 people. The great news is that purpose built devices like OrCam MyEye, Envision Glasses are already serving this purpose in a safe way. These devices process visual information on-device to describe a scene or read text aloud, and are not built around persistent recording, cloud storage, or social engagement.
Smart glasses on the other hand do precisely that. Meta’s own 2025 policy update says that they store photos/video on the paired phone and keep voice recordings in the cloud for up to a year by default.
Experts in need of hands free recording are already served by head mounted portable cameras.
On the other hand, 50.38% of Australia’s 27.8 million population is female. If we restrict this analysis to women aged 15+, most prone to experiencing tech based violence, that’s roughly 11.6 million.
Applying government established rates of documented harm and unlike benefits choosing the conservative estimates.
15% of women have faced image based abuse.
31% have faced tech based abuse.
Rise in misogyny, showing 78% of 2,600 teachers reported seeing or experiencing gendered violence at school.
With some overlap that is conservatively between 1.7 and 2.7 million women and girls. Roughly 2 to 3 times the size of the vision impaired benefit group who already have purpose built, better designed alternatives.
On its benign surface the right to erasure sounds perfect.
If somebody records you? Simple, fill a form.
The company promises that in 30 days this data will be deleted.
We believe them.
But, should we?
In 2023, US FTC vs Amazon Alexa. Amazon told parents they could delete their children’s voice recordings. The FTC found Amazon deleted from some databases and kept the transcripts in others, and continued using the retained data to train Alexa’s algorithm. Amazon was fined $25 million and given an order specifically barring it from using data subject to deletion requests to build future products.
In 2025 Apple settled a lawsuit for $95 million over Siri recordings, sometimes captured without the wake word, being manually reviewed by contractors without telling its customers. This included accidentally captured private medical, legal, and personal conversations. The original alarm was raised in 2019 by a whistleblower. Apple did not admit to any wrongdoing.
In 2019 Meta received the largest privacy penalty in FTC history, $5 billion, specifically for deceiving users about how much control they actually had over their data and who could access it, allowing Cambridge Analytica to harvest data from tens of millions of users for political targeting.
We are all believing the word of entities known to lie for profit.
That’s not the worst part though. The EU and UK do have codified erasure laws which give companies up to 90 days to delete customer data.The thing is, they don’t really have a mechanism to ensure this happens beyond believing internal records, audit logs and correspondence. No regulator actually verifies this at a systems level.
This is before AI comes into the picture, because while organisations have spent over $1.37 trillion this year alone investing in AI, a field that stays unexplored is data unlearning. This means that consent or no, once our data has been fed into an AI system that derives patterns from it, it’s currently impossible to truly delete this data at LLM scale.
The profit over people cycle
Two recent examples of overly permissive, low social utility products provide the perfect analogy to what can go wrong.
Gambling apps that were allowed to operate indiscriminately are now under scrutiny. Going through expensive parliamentary hearings and litigation that are woefully ineffective in preventing the widespread harm they have been proven to cause, almost $32 billion in individual finances alone. The 2023 inquiry, You Win Some You Lose More, made 31 recommendations. None had been implemented two years later.
After the fact opt out policies and little regulation over advertising bans do too little too late. BetStop, the national self-exclusion register, had just 30,032 active exclusions as of June 2025. That’s against 3.1 million Australians, 15% of adults, who report gambling related harm each year.
Indiscriminate social media use that has proven to affect mental health and well being is now being banned for children under 16. In reality, the mental health harms are not limited to children, only more pronounced and studied. If anything the evidence cuts the other way. A 2025 meta-analysis found the effect sizes for social media and depression are generally larger in adults than adolescents.
The ban has done little to limit use of social media by children. A University of Newcastle study found over 85% of under-16s were still using the covered platforms three months after the ban took effect, mostly by simply self-declaring their age. eSafety’s own research found only about a third of young children with existing accounts actually stopped using them.
This is what happens when post hoc fixes are the default mode of operation versus pragmatic prevention.
The right to not be recorded
What we already have, and should codify, is the right not to be recorded. Not by an opt out, by default.
This means, on mobiles or mass recording devices, no data leaves the device without explicit removal of human faces. Edge computing ensures that erasure happens at source, at the time of collection versus months later on a server after a request.
The removal algorithms should be designed in partnership with trusted organisations like the CSIRO, that ensure that the government has a feed of traceable & auditable logs of information that goes to servers as metadata. Detailing precisely what has been sent to the cloud. Accessible publicly through a dedicated public register, not an FOI request queue running at 74% on time compliance with a growing backlog, where organisations that ensure safety and privacy can query this information to see exactly what type of data was sent to a vendor and when.
It is time the government protects rights we already have instead of trying to replace them with watered down consolation prizes that strip us of our agency and volition.
Common sense in action
Watching our government, on both sides of the political debate, continue to tout common sense without exercising it is exhausting.
Public theatre and long drawn out, tax funded inquiries that produce recommendations which never get adopted is precisely the pattern that big tech and corporations count on. Individual litigation over social prevention is exactly the grey area that creates a world where we need to provide sexual strangulation explanations to our underage daughters and consent education to grown men.
The failure mode is no longer just a few harmless images circulated among friends. In the coming decades, it could be thousands of young men recording themselves performing life threatening stunts, hundreds of young girls resorting to self harm from tech based abuse, dozens of people mistakenly penalised due to algorithmic misidentification. Years later, we have another royal commission where we hear impact statements, receive yet another list of recommendations that get pruned, debated and hollowed out.
All the while, the fact remains that the utility case for smart glasses is narrow and already being better served by purpose built devices. The risk case is broad and defers responsibility to people who never opted into anything, and for whom no erasure mechanism exists even in principle, because current erasure law is written around the wearer/user, not the bystander.
That is the real asymmetry. A right to erasure, however well designed, protects the person wearing the glasses.
Banning these devices doesn’t disadvantage anyone except the companies selling them, while protecting the general public. That is putting people over profit.
This isn’t a foregone conclusion, yet. Australia’s Tranche 2 privacy reform consultation (which includes the right-to-erasure proposal) closes September 18. This is our opportunity to send a loud and clear message to the government, that we are no longer accepting the bare minimum.
Take 2 minutes in your day to add your objection to smart glasses and call for the right not to be recorded versus the right to be erased. Share it with as many people as you can while we still can.