[GistNoesis]

One obvious problem with human review is steganography.

The picture can look normal to the human eye, but if it contains hidden content (in the least significant bit of each pixel for example so that the hash is unchanged), a forensic software will definitely notice, raise some flags, and extract the hidden offensive content automatically, leaving the human reviewer no other choice but to report you.

If Apple says they are not going to look for hidden content, then they are just handling a free pass which render the whole scanning thing pointless.

[simondotau]

I'm confused what scenario you're positing here. Given the widespread adoption of encrypted communications, steganography is of no use to traffickers of CSAM. Steganography generally serves only one purpose, which is to transfer material in public view with plausible deniabilty—such as leaking material out of a military facility which has exceedingly robust data protection processes.

Apple have explicitly said that their hash algorithm is only concerned with visible elements of the image.

[GistNoesis]

I'm speaking about the adversarial scenario of an attacker trying to frame a target. He just need to get on your phone an image with hidden content that has a hash collision with the database.

Traffickers and consumers of CSAM know that their content is illegal to possess and store so they sometime use steganography software to store the offensive data inside their innocuous photo library. This way when they can browse their private collection via the lens of the steganography software and they don't have some suspicious encrypted file that would attract attention of someone they share the computer with.

[simondotau]

You seem to be confused. As you said yourself, steganographic concealment would, by its very nature, not change the perceptual hash of the visible image. If the visible image doesn't match an known hash, the steganographically modified version isn't going to either.

[GistNoesis]

This sit on top of the perceptual hash collision.

First you generate an innocuous image that has a bad hash collision. (This is easy because perceptual hash are not cryptographically secure). Then in a second step you hide some offensive content in it via steganography without changing the hash. Then you send the image to the target.

He stores it in his cloud, it gets flagged because of the hash collision, so it get a manual review. The manual review take the image through some forensic software, which will catch the steganography (because the attacker will have chosen a weak scheme) which will reveal the hidden offensive content and then report you.

[simondotau]

The manual review process only involves a severely transformed (low resolution, greyscale) version of the image which is attached to the safety token. The ability to decrypt any original files only occurs if the human review process confirms the presence of CSAM.

[GistNoesis]

I don't have a lot of info on the quality of the visual derivative.

But since a human should look at it should have enough details to distinguish subtle cases like the age of the people in the picture, otherwise it's even more concerning.

If some human has enough info to make this call then the low-res greyscale visual derivative should still raise some flags if it get through a forensic software, as steganography software usually offer some resistance against usual compression artifacts.

[madmoose]

What you're talking about here has nothing to do with what Apple is implementing.