back to article Stanford Internet Observatory raises alarm over 'serious failings with the child protection systems at Twitter'

Twitter failed to take down numerous images of child sexual abuse material posted on its platform over two months, researchers at the Stanford Internet Observatory will allege in an upcoming report. The academic group said it was looking at the wider issue of child exploitation online when it "discovered serious failings with …

  1. Ken Y-N

    The subhead says: "Researchers find 100,000 accounts spamming child abuse material"

    I don't see that supported in the article or in the links - they only seem to have checked 100,000 tweets and found 40 dodgy image hash matches. Still far too many, of course.

    1. Anonymous Coward
      Anonymous Coward

      And also what is the false positive ratio with this Microsoft hashing algorithm?

      1. tiggity Silver badge

        @AC - indeed, false positives occur and relevant to this article, especially as (for many obvious reasons) they do not want to view the hash matched images.

        Not sure about how many hashes CSAM uses, but that's also relevant.

        .. and lets not even get started on false positive paradoxes.

        But obviously, by nature of the potential content, this area difficult to assess for researchers - I hope they did pass on details of the images to those in law enforcement who are legally able to assess such images (vile job to have) and prosecute if genuine nasties found.

        1. Michael Wojcik Silver badge

          Not sure about how many hashes CSAM uses, but that's also relevant.

          CSAM doesn't use any hashes. PhotoDNA computes hashes of images; the hashes aren't part of the image. I have no idea why you think that's "relevant".

          Collisions exist, of course, as they do with any non-perfect hash, by the Pigeonhole Principle. And PhotoDNA improves recall (reduces false negatives) by deliberately discarding some information which is often altered when sharing photos; its hash process converts the image to greyscale and quantizes the image. That increases the probability of a collision.

          But collisions are still pretty fucking unlikely in actual real-world use. PhotoDNA uses Sobel gradients computed on overlapping 6x6 greyscale-pixel grids. Like other perceptual image hashes, its precision is going to be strongly influenced by human-perceptible differences in images, such as the number, shape, and orientation of objects in the image. And the resulting hash is 1152 bits, so the probability of a random collision is nearly 1 in 17179869184. If you have seventeen billion images in your Twitter feed, it might be something to worry about.

    2. Killfalcon

      I parsed that as 100,000 tweets of CSAM, but only 40 different images being used. It's not very clearly written.

  2. Anonymous Coward
    Anonymous Coward

    867,000,000 / 100,000 * 40 = 346,800

    Twitter published about ~ 867 million tweets daily.

  3. Neil Barnes Silver badge

    Problem: researchers find too many child abuse posts

    Solution: make access too expensive so they can't...

    1. MiguelC Silver badge

      Re: Problem: researchers find too many child abuse posts

      They might call it "collateral bonus"

    2. Anonymous Coward
      Anonymous Coward

      Re: Problem: researchers find too many child abuse posts

      Oh, you got there first. Was going to say that this...

      > criticized Twitter for raising prices, and said doing so makes it more difficult for academics to study the platform and hold it accountable

      ...sounds like it's a "feature" not a bug.

      Not that I'm suggesting Elon "Pedo Guy" Musk is a literal pedo (*), just that it suits him not to get into trouble for other people posting such things he clearly doesn't care about having Twitter fix under his leadership.

      Indeed, his mockery of critics pointing out the effects of his sacking most of the staff who deal with stuff like this also implies that he's not bothered by people posting such things on his platform so long as he doesn't get into trouble for it.

      (*) Musk's defense team already won that libel case on the basis that he simply the other guy was a "creepy" person (*cough*bullshit*), so we can say the same thing Musk.

  4. Pascal Monett Silver badge

    As if Musk cares

    So the lowest tier is $42K ? Guess what, he already doesn't pay his bills and he's not known for doing favors.

    Twitter is dead. Leave it to rot.

    1. Disgusted Of Tunbridge Wells Silver badge

      Re: As if Musk cares

      You know this definitely pre-dates Musk, right?

      And you know that pre-Musk, Twitter was brazenly engaging in election interference.

      Just because they were covering for "your side" doesn't make it good.

      1. Disgusted Of Tunbridge Wells Silver badge
        WTF?

        Re: As if Musk cares

        5 people who think that corporations suppressing speech and lying in order to win an election is great as long as it's their side doing the lying and suppressing.

        1. sabroni Silver badge
          FAIL

          Re: 5 people who think that corporations suppressing speech ... is great

          Or one uneducated bigot who has no idea what's going on at Twitter?

          https://www.techdirt.com/2023/06/05/twitter-admits-in-court-filing-elon-musk-is-simply-wrong-about-government-interference-at-twitter/

          Never mind the source, you can click through to the filing from Twitters lawyers where they explain why "the Twitter Files" don't show what moron's think they do.

          1. Anonymous Coward
            Anonymous Coward

            Re: 5 people who think that corporations suppressing speech ... is great

            Twitters lawyers will do ANYTHING to divert something this big from them. They know full well that if they fail they will never work in tech again and also that other platforms such as facebook will then be under scrutiny.

            The simple fact is the feds and various govt persons and agencies had a direct route into twitter and skirted very close to (if not actually over) the edge of violating some fundamental laws of the USA.

            However these so called 'journalists' are happy as the people silenced were thought criminals so it was all just fine.

            If it had been the Trump white house getting pantifa or BLM banned from twitter this would look very different.

            1. Ace2 Silver badge

              Re: 5 people who think that corporations suppressing speech ... is great

              Simple facts… in your simple mind.

      2. Ace2 Silver badge

        Re: As if Musk cares

        Oh Disgusting, still at it, I see.

        WTF is the matter with you? Are you being paid to not understand anything about these issues?

        1. Michael Wojcik Silver badge

          Re: As if Musk cares

          Easy, now. Accusing posters of shilling is a violation of the Reg forum terms of service, and we should give DoTW the benefit of the doubt. It's entirely possible he's just refusing to understand for his own amusement.

  5. Anonymous Coward
    Facepalm

    Yet more Twitter trashing

    “Having no remaining Trust and Safety contacts at Twitter”

    Have they tried DMing Elon Musk?

    What other platform have these “researchers” performed such tests on?

    1. IGotOut Silver badge

      Re: Yet more Twitter trashing

      "Having no remaining Trust and Safety contacts at Twitter”

      "Have they tried DMing Elon Musk?"

      Is Elon now personal checking content? Has he started engaging with people on a professional basis that may raise criticism of his management style?

      "What other platform have these “researchers” performed such tests on?"

      I don't know, maybe you could read the article to see they have and the info is to follow?

      And not sure why you put researchers I quotes. Maybe if you got off your Elin feed on twitter, you could "research" who they are. Tell you what, let me spoon feed you.

      https://pit.stanford.edu/university-organizations/internet-observatory/

      1. dc_m

        Re: Yet more Twitter trashing

        I was wondering the same thing. If Stanford is not the definition of a research establishment, then what is it?

        1. Anonymous Coward
          Anonymous Coward

          Re: Yet more Twitter trashing

          https://www.wsj.com/articles/instagram-vast-pedophile-network-4ab7189

          Hmm... this can't be good.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like