They were using photoDNA and Youtube CSAI Match before the article. This is the same as what Reddit uses. I see it mentioned here back in March https://www.reuters.com/article/idUSKBN21D3E9