In the past, when someone tipped off the Internet Watch Foundation’s (IWF) criminal content reporting hotline to an online video they thought included child sexual abuse material, an analyst at the U.K. nonprofit often had to watch or fast forward through the entire video to investigate it.
Because people sharing videos of child sexual abuse often
embed this illegal content in an otherwise innocuous superhero flick,
cartoon or home movie, it could take 30 minutes or several hours to find
the content in question and determine whether the video should be taken
down and reported to law enforcement.
Last year, IWF, a global watchdog organization, started leveraging
PhotoDNA — a tool originally developed by Microsoft in 2009 for still
images — to identify videos that have been flagged as child
sexual abuse material. Now it often takes only a minute or two for an
analyst to find illegal content.
Microsoft is now making PhotoDNA for Video available for free, and
any organization worldwide interested in using the technology can visit
the Microsoft PhotoDNA website to find out more, or to contact the team.
“It’s made a huge difference for us. Until we had PhotoDNA for Video,
we would have to sit there and load a video into a media player and
really just watch it until we found something, which is extremely
time-consuming,” says Fred Langford, deputy chief executive of IWF,
which collaborates with sexual abuse reporting hotlines in 45 countries
around the world.
“This means we can identify and disrupt online sexual abuse and help victims much faster,” says Langford.
“We don’t want this illegal content shared on our products and services. And we want to put the PhotoDNA tool in as many hands as possible to help stop re-victimization.”
Courtney Gregoire, Microsoft Digital Crimes Unit
PhotoDNA for Video builds on the same technology employed by PhotoDNA,
a tool Microsoft developed with Dartmouth College that is now used by
over 200 organizations around the world to curb sexual exploitation of
children. Microsoft leverages PhotoDNA to protect its customers from
inadvertently being exposed to child exploitation content, helping to
provide a safe experience for them online.
PhotoDNA has also enabled content providers to remove millions
of illegal photographs from the internet; helped convict child sexual
predators; and, in some cases, helped law enforcement rescue potential
victims before they were physically harmed.
In the meantime, though, the volume of child sexual exploitation
material being shared in videos instead of still images has ballooned.
The number of suspected videos reported to the CyberTipline
managed by the National Center for Missing and Exploited Children
(NCMEC) in the United States increased tenfold from 312,000 in 2015 to
3.5 million in 2017. As required by federal law, Microsoft reports all
instances of known child sexual abuse material to NCMEC.
Microsoft has long been committed to protecting its customers from
illegal content on its products and services, and applying technology
the company already created to combating this growth in illegal videos
was a logical next step.
“Child exploitation video content is a crime scene. After exploring
the development of new technology and testing other tools, we determined
that the existing, widely used PhotoDNA technology could also be used
to effectively address video,” says Courtney Gregoire, Assistant General
Counsel with Microsoft’s Digital Crimes Unit. “We don’t want this
illegal content shared on our products and services. And we want to put
the PhotoDNA tool in as many hands as possible to help stop the
re-victimization of children that occurs every time a video appears
A recent survey of survivors of child sexual abuse
from the Canadian Centre for Child Protection found that the online
sharing of images and videos documenting crimes committed against them
intensified feelings of shame, humiliation, vulnerability and
powerlessness. As one survivor was quoted in the report: “The abuse
stops and at some point also the fear for abuse; the fear for the
material never ends.”
The original PhotoDNA helps put a stop to this online
recirculation by creating a “hash” or digital signature of an image:
converting it into a black-and-white format, dividing it into squares
and quantifying that shading. It does not employ facial recognition
technology, nor can it identify a person or object in the image. It
compares an image’s hash against a database of images that watchdog
organizations and companies have already identified as illegal. IWF,
which has been compiling a reference database of PhotoDNA signatures,
now has 300,000 hashes of known child sexual exploitation materials.
PhotoDNA for Video breaks down a video into key frames and
essentially creates hashes for those screenshots. In the same way that
PhotoDNA can match an image that has been altered to avoid detection,
PhotoDNA for Video can find child sexual exploitation content that’s
been edited or spliced into a video that might otherwise appear
“When people embed illegal videos in other videos or try to hide them
in other ways, PhotoDNA for Video can still find it. It only takes a
hash from a single frame to create a match,” says Katrina Lyon-Smith,
senior technical program manager who has implemented the use of PhotoDNA
for Video on Microsoft’s own services.
Organizations that are already using an on-premise
version of PhotoDNA to remove illegal images will be able to seamlessly
add the capability to identify videos. Microsoft is also looking for
partners to test the video technique on its PhotoDNA Cloud Service.
Automated tools like PhotoDNA have made a huge difference in the
fight against online child exploitation, particularly for smaller
companies that otherwise wouldn’t have the capacity or know how to find
illegal content on their apps and websites, says Cecelia Gregson, a
senior King County prosecutor and attorney for the Washington Internet
Crimes Against Children Task Force.
Gregson estimates that 90 percent of the
cases she investigates now come from CyberTipline reports submitted by
companies using PhotoDNA to keep their platforms clean. Under federal
law, all internet and email service providers are required to report
knowledge of child pornography to NCMEC.
“It’s made a huge difference…We can identify and disrupt online sexual abuse and help victims much faster.”
Fred Langford, Internet Watch Foundation
“This is not about looking at someone’s online shopping patterns or uploaded family photos. We are seeking files depicting the sexual abuse of children,” says Gregson. “We are concerned with protecting child victims, and about making sure the places you go online and your children go online are not riddled with images of child abuse and exploitation. The technology can also help us identify child sexual predators whose collections of images can cause further psychological, emotional and mental trauma to their victims.”
Since PhotoDNA and other tools became widely available, the number of reports to NCMEC’s CyberTipline
has grown from 1 million in 2014 to 10 million in 2017, says John
Shehan, vice president for NCMEC’s exploited children division.
“These technologies allow companies, especially the hosting
providers, to identify and remove child sexual content more quickly,”
says Shehan. “That’s a huge public benefit.”
Internet service providers may have better success at scanning their networks to actively seek out illicit images of child abuse, thanks to technology donated by Microsoft and Dartmouth College.
On Wednesday, the software giant and the well-known college announced that they had developed a software program to match modified images to the original by using a form of robust hashing that can ignore certain types of changes, such as resizing, cropping and the inclusion of text. The team donated the program, dubbed PhotoDNA, to the National Center for Missing and Exploited Children.
The NCMEC will make the program available to ISPs to detect the “worst of the worst” in child pornography — those images that show pre-pubescent children being sexually abused, said Ernie Allen, CEO and president of the NCMEC.
The intent is to “use the technology very narrowly and very specifically,” Allen said.
The agreement follows a number of other successful initiative in fighting child abuse online. In June 2008, three ISPs signed an agreement with the New York State Attorney General’s office to police their networks for child pornography and donate money to the state and the NCMEC to fund investigations. In 2007, MySpace agreed with the attorneys general of more than 40 states to turn over information regarding sex offenders on its network.
While law enforcement has successfully prosecuted hundreds of cases of possession and distribution of illicit images, a small number of cases have underscored overzealous prosecutions. In one case, a Massachusetts government agency fired and reported one of its workers for having child pornography on his laptop, but a later investigation showed that the lack of functioning antivirus software resulted in his laptop being compromised and subsequently filled with illicit images.
Microsoft has already tested the software on its networks and plans to roll out the tool to scan public sources for images for child pornography, said Brad Smith, senior vice president and general counsel at the software giant.
“It is not enough to catch the perpetrators, we have to stop the images to prevent the subjects from being a victim again,” Smith said.
While Microsoft will scan public sources for matches to a small database of the worst abuse images, the software giant will not scan private data nor communications, Smith said. ISPs, the government and privacy advocates should discuss the legal and policy issues of such scanning, he said.
Child pornography is a major priority of law enforcement and the detection of images of abuse has grown significantly, according to the NCMEC. Since 2003, the organization has viewed and analyzed 30 million images classified as child pornography, the group claims. Allen predict that the group will deal with another 9 million in 2010.
Much of the increase in child pornography is due to the Internet’s ability to allow communities to form among traders of child pornography, he said.
“They (the criminals) no longer view themselves as aberrant,” Allen said. “We made enormous progress on the commercial side … but it has migrated to the noncommercial side.”
In the latest announcement, a large scale test of the PhotoDNA tool found that less than one false positive occurred in every billion images scanned, said Hany Farid, a professor of computer science at Dartmouth and co-developer of PhotoDNA. In addition, the software recognizes about 98 percent of images derived from those in its database.
“We tested it over billions and billions of images,” he said. “We tried very hard to make it very efficient … and to minimize the false alarm rate.”