4 ms·

> two intersection of two independent CSAM databases. I don't think that offers meaningful protection. A malicious government could acquire or produce real CS

> two intersection of two independent CSAM databases. I don't think that offers meaningful protection. A malicious government could acquire or produce real CSAM, or at least a convincing fake of it, then modify that image subtly to match the hash of the political content they were targeting (I believe such engineered collisions have already been demonstrated.) They would then be able to get that manipulated CSAM into as many independent CSAM databases as they wanted, simply by uploading it to the web then reporting it to all those independent databases.

quitit 5y ago
This isn't feasible for a variety of reasons: 1. Forced hash collisions are fragile in comparison to the genuine material. CSAM hashing is devised in a way to allow change in the image without significant variation to the hash. Forced collisions however don't possess that property as their true hash is different. 2. It's trivial to detect images with false hashes, and different hashing methods will not produce collisions. 3. The task you're proposing, in itself is not pragmatic nor effective: dissenting political materials are easily created fresh (i.e. no colliding hash) and exists in many forms that go well beyond an image. 4. Human reviewers again would see the pattern and adapt the system accordingly 5. CSAM database holders aren't idiots. They understand the significance of their hash libraries and work against manipulation from CSAM peddlers.
- BrazzVuvuzela 5y ago
  Points 1 and 3 assume any government wouldn't choose to target people who had a specific document, I think this is a poor assumption. Hash fragility doesn't seem relevant if both the hashing algorithms and the target content are known to the government Point 2 and 5 I don't understand, could you expand on it? If the government produced manipulated CSAM that didn't exist in any other pre-manipulation form, how would the manipulation be detected? And supposing it were detected, would CSAM database holders choose not to add that image to their databases just because it was manipulated? If that were the case, it would give pedophiles a way to exempt their images from the database too. Point 4 is neutralized by the government demanding local human reviewers for phones located in their country. China already does this with iCloud, they could demand the same for the CSAM reviewers, then coerce those reviewers or simply select them for party loyalty in the first place.

quitit 5y ago

This isn't feasible for a variety of reasons: 1. Forced hash collisions are fragile in comparison to the genuine material. CSAM hashing is devised in a way to allow change in the image without significant variation to the hash. Forced collisions however don't possess that property as their true hash is different. 2. It's trivial to detect images with false hashes, and different hashing methods will not produce collisions. 3. The task you're proposing, in itself is not pragmatic nor effective: dissenting political materials are easily created fresh (i.e. no colliding hash) and exists in many forms that go well beyond an image. 4. Human reviewers again would see the pattern and adapt the system accordingly 5. CSAM database holders aren't idiots. They understand the significance of their hash libraries and work against manipulation from CSAM peddlers.

BrazzVuvuzela 5y ago
Points 1 and 3 assume any government wouldn't choose to target people who had a specific document, I think this is a poor assumption. Hash fragility doesn't seem relevant if both the hashing algorithms and the target content are known to the government Point 2 and 5 I don't understand, could you expand on it? If the government produced manipulated CSAM that didn't exist in any other pre-manipulation form, how would the manipulation be detected? And supposing it were detected, would CSAM database holders choose not to add that image to their databases just because it was manipulated? If that were the case, it would give pedophiles a way to exempt their images from the database too. Point 4 is neutralized by the government demanding local human reviewers for phones located in their country. China already does this with iCloud, they could demand the same for the CSAM reviewers, then coerce those reviewers or simply select them for party loyalty in the first place.