PhotoDNA
A hash-matching technology developed by Microsoft to detect and track child sexual abuse material (CSAM), identifying illegal content by comparing hash values against a database of known violations.
Core Facts
Timeline (last 90 days)
Generative AI models can be induced to generate entirely fabricated abuse images (synthetic CSAM), and such content does not match any known hash database, defeating traditional PhotoDNA-style matching.
Content fingerprinting (hash matching) technologies (such as PhotoDNA) compute cryptographic hashes of known illegal images or videos and then compare them against content uploaded to platforms, enabling identification of illegal files that are repeatedly circulated without viewing the original content.
Hash matching (such as PhotoDNA) is the mainstream technology for platforms to detect CSAM, computing perceptual hashes of known illegal images and comparing them against uploaded content.
Technically, CSAM detection relies mainly on two approaches: hash matching against known illegal content (such as perceptual hashing technologies like PhotoDNA) and classification and identification based on machine learning models.
All Facts (4)
Hash matching (such as PhotoDNA) is the mainstream technology for platforms to detect CSAM, computing perceptual hashes of known illegal images and comparing them against uploaded content.
65%UnverifiedGenerative AI models can be induced to generate entirely fabricated abuse images (synthetic CSAM), and such content does not match any known hash database, defeating traditional PhotoDNA-style matching.
50%UnverifiedContent fingerprinting (hash matching) technologies (such as PhotoDNA) compute cryptographic hashes of known illegal images or videos and then compare them against content uploaded to platforms, enabling identification of illegal files that are repeatedly circulated without viewing the original content.
50%UnverifiedTechnically, CSAM detection relies mainly on two approaches: hash matching against known illegal content (such as perceptual hashing technologies like PhotoDNA) and classification and identification based on machine learning models.
50%