Paul Moore - Security Consultant 

#HMDFuse: Sophisticated #AI, #CSAM detection, can't be disabled or circumvented.

"A phone incompatible with pornography", they said.

Turn it sideways.
paul.reviews/harmblock-worl…
#childSafety #harmblock #AI #HMD #xplora #SafeToNet @NCSC @CEOPEducation @IWFhotline

Let's explore @LordJohnNash's comments on the unforgiveable intrusion into our everyday lives...

"This is neither surveillance nor a back door"
It's literally watching everything you do for the purposes of preventing crime; that is the definition of surveillance.

"nothing is sent to anyone"
This is *almost* impossible to achieve in the real world.

Even if you never intentionally send reports, data is often collated & stored in a manner which can later imply or directly demonstrate the user viewed harmful content.

"It simply stops the image appearing..."
The only representative model, HarmBlock (which they conveniently dropped from this announcement), was so fragile, simply rotating the device rendered the "protection" moot.

"82% of people who searched for abuse and were warned, never searched again"
I call BS. Whilst the stats are likely accurate, they're almost certainly confined to the same site. If you search Google for "fridge freezers" and got 0 results, of course you're unlikely to use the same search term again. You will, however, almost certainly use another site.

"The software is trained on exactly those pictures"
Which images? Throughout every publication, they regularly conflate "nudity" with "CSAM" material. They require two entirely different training models.

We can test a model for nudity quite easily. But, only @IWFhotline can test for CSAM efficacy legally.

Regardless, conflating those categories at this early stage is a recipe for disaster.

"It learns the difference..."
No, it doesn't. To learn from your mistakes, you need training data; data which supposedly never leaves the device. It can only use external data to improve the model... and who's the arbiter of such information?

Updates & training can change behaviour, but it's not "learning" in the traditional sense. If your phone blocks a false positive today, it'll block the same false positive tomorrow.

That picture of your elongated cat? Blocked.(paul.reviews/harmblock-worl…)

"Improves with every update..."
... and therein lies the problem.

One update and suddenly, it's no longer isolated to CSAM or nudity. It'll be gore, harmful speech, criticism of government policy.

Someone unknown to you can control/manipulate everything you see.

"because nothing leaves the device, a wrong call can never become a report or an allegation"
So millions of devices can potentially flag CSAM material and, by design, nobody is notified?

It can't prevent harm from happening & doesn't notify anyone if/when it's shared. Of course, some will argue that if fewer people can view it, demand will drop.

That may be true, but it's more likely to drive this abhorrent behaviour underground to platforms over which they have no control.

They know this.

Everyone involved in this discussion, myself included, labelled this an escalation path and that's already been proven.

"It's just social media..."
"It's just social media and gaming platforms..."
"It's just social media, gaming platforms and auction sites..."
"Throw in Spotify too"

... all in the space of a few months. This isn't censoring access to platforms... it runs much deeper. This is censoring basic communication, regardless of platform.

The foundations for wider surveillance is an unfortunate & entirely unforeseen consequence of "protecting the children"

This is massive government overreach, nothing more.

When responsible firms like @Apple pushed back against their intrusion, they didn't wave the white flag. They regrouped, changed their approach and sold this whole debacle as a way to protect children in order to provoke a different response.

Now, the criticism isn't against retaining #privacy. It's "if you don't want to protect children, you're on the side of predators".

That's a much easier sell. If you fall for it.

telegraph.co.uk/politics/2026/…

Received an email purportedly from @AmazonUK?

You can now verify it's real by forwarding it to verify@amazon.com

Now if only it worked...

I suspect my name is mud in this discussion.

My research destroyed both the NPCC's #cyberAlarm and #harmblock yet they still insist on pushing tech they don't understand.
observer.co.uk/news/national/…

Gov't: "Forcing regular #password changes does more harm than good"

Also gov't...