US AGs: We need law to purge the web of AI-drawn child sex abuse material
(2023/09/06)
- Reference: 1693958524
- News link: https://www.theregister.co.uk/2023/09/06/ai_csam_national_law_call/
- Source link:
The National Association of Attorneys General, the body that all US states and territories use to collaboratively address legal issues, has urged Congress to pass legislation prohibiting the use of AI to generate child sex abuse images.
In a letter
Advances in generative AI technology have made it easy to create realistic images that depict real people in highly compromising or disturbing made-up scenarios, or fake people in fictitious circumstances. Online safety groups and law enforcement agencies have [2]noticed an increase in these so-called deepfakes, images or videos in which a real person's face is pasted on someone else's body to produce fake content. Deepfakes of children's photos can be used to tweak existing CSAM to churn out more of that vile content.
[3]
Online text-to-image tools can also fabricate CSAM that looks realistic but does not depict an actual child. Although companies operating text-to-image services and software have strict policies and often block images containing nudity, users can sometimes find ways to bypass restrictions. Open-source models can also generate CSAM and are harder to police as they can be run locally.
[4]
[5]
Creating pornographic deepfakes depicting real people is illegal in at least some parts of the United States. Earlier this year, prosecutors in Long Island, New York, charged a man for creating and sharing sexually explicit deepfakes depicting "more than a dozen underage women," using images he took from social media profiles. This machine-made material was shared on porno sites along with the victims' personal information and calls for fellow perverts to harass them. The 22-year-old man was [6]sentenced to six months in prison and given ten years’ probation with significant sex offender conditions.
However, no federal legislation prohibits making NSFW deepfakes without consent. The laws are murkier when it comes to completely fake AI-generated CSAM, in which the victims are not real people.
[7]
The National Association of Attorneys General argued that such material is not victimless, as tools capable of generating such images was likely trained on actual CSAM, the creation of which harmed actual children. Making more entirely virtual AI could therefore fuel further child exploitation and spread more revolting and illegal content online.
The AGs therefore want laws or other tools to combat deepfake porn, whether it's of real people manipulated into fake situations without permission, or totally fake stuff that was likely developed from actual illegal material.
[8]Deepfakes being used in 'sextortion' scams, FBI warns
[9]China bans deepfakes created without permission or for evil
[10]Scanning phones to detect child abuse evidence is harmful, 'magical' thinking
[11]Police lab wants your happy childhood pictures to train AI to detect child abuse
"One day in the near future, a child molester will be able to use AI to generate a deepfake video of the child down the street performing a sex act of their choosing," Ohio's Attorney General Dave Yost said in a statement. "Graphic depiction of child sexual abuse only feeds evil desires. A society that fails to protect its children literally has no future," he said.
The letter was spearheaded by South Carolina's Attorney General Alan Wilson, [12]according to the AP.
"First, Congress should establish an expert commission to study the means and methods of AI that can be used to exploit children specifically and to propose solutions to deter and address such exploitation," the document states.
[13]
"Second, after considering the expert commission's recommendations, Congress should act to deter and address child exploitation, such as by expanding existing restrictions on CSAM to explicitly cover AI-generated CSAM. This will ensure prosecutors have the tools they need to protect our children."
Addressing fake CSAM is tricky. Typical techniques that detect the illegal content relies on hashing known images that are being circulated online. It’s therefore difficult to identify new images, especially if they have been doctored using software. The Attorneys General, however, believe lawmakers must act because the technology will continue to evolve.
"We are engaged in a race against time to protect the children of our country from the dangers of AI. Indeed, the proverbial walls of the city have already been breached. Now is the time to act," the letter concludes. ®
Get our [14]Tech Resources
[1] https://regmedia.co.uk/2023/09/05/handout_ag_letter_csam.pdf
[2] https://www.theregister.com/2023/06/08/ai_deepfakes_sextortion_fbi/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.nassauda.org/CivicAlerts.aspx?AID=1512
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2023/06/08/ai_deepfakes_sextortion_fbi/
[9] https://www.theregister.com/2022/12/12/china_deep_synthesis_deepfake_regulation/
[10] https://www.theregister.com/2022/10/13/clientside_scanning_csam_anderson/
[11] https://www.theregister.com/2022/06/03/police_australia_ai/
[12] https://apnews.com/article/ai-child-pornography-attorneys-general-bc7f9384d469b061d603d6ba9748f38a
[13] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[14] https://whitepapers.theregister.com/
In a letter
[1]PDF
to the leaders of the Senate and the House of Representatives on Tuesday, the Attorneys General requested lawmakers appoint an expert to study how the content-making machine-learning technology can be used to exploit children, with the goal of establishing new laws, rules, or regulations to protect against AI-generated child sexual abuse material (CSAM).Advances in generative AI technology have made it easy to create realistic images that depict real people in highly compromising or disturbing made-up scenarios, or fake people in fictitious circumstances. Online safety groups and law enforcement agencies have [2]noticed an increase in these so-called deepfakes, images or videos in which a real person's face is pasted on someone else's body to produce fake content. Deepfakes of children's photos can be used to tweak existing CSAM to churn out more of that vile content.
[3]
Online text-to-image tools can also fabricate CSAM that looks realistic but does not depict an actual child. Although companies operating text-to-image services and software have strict policies and often block images containing nudity, users can sometimes find ways to bypass restrictions. Open-source models can also generate CSAM and are harder to police as they can be run locally.
[4]
[5]
Creating pornographic deepfakes depicting real people is illegal in at least some parts of the United States. Earlier this year, prosecutors in Long Island, New York, charged a man for creating and sharing sexually explicit deepfakes depicting "more than a dozen underage women," using images he took from social media profiles. This machine-made material was shared on porno sites along with the victims' personal information and calls for fellow perverts to harass them. The 22-year-old man was [6]sentenced to six months in prison and given ten years’ probation with significant sex offender conditions.
However, no federal legislation prohibits making NSFW deepfakes without consent. The laws are murkier when it comes to completely fake AI-generated CSAM, in which the victims are not real people.
[7]
The National Association of Attorneys General argued that such material is not victimless, as tools capable of generating such images was likely trained on actual CSAM, the creation of which harmed actual children. Making more entirely virtual AI could therefore fuel further child exploitation and spread more revolting and illegal content online.
The AGs therefore want laws or other tools to combat deepfake porn, whether it's of real people manipulated into fake situations without permission, or totally fake stuff that was likely developed from actual illegal material.
[8]Deepfakes being used in 'sextortion' scams, FBI warns
[9]China bans deepfakes created without permission or for evil
[10]Scanning phones to detect child abuse evidence is harmful, 'magical' thinking
[11]Police lab wants your happy childhood pictures to train AI to detect child abuse
"One day in the near future, a child molester will be able to use AI to generate a deepfake video of the child down the street performing a sex act of their choosing," Ohio's Attorney General Dave Yost said in a statement. "Graphic depiction of child sexual abuse only feeds evil desires. A society that fails to protect its children literally has no future," he said.
The letter was spearheaded by South Carolina's Attorney General Alan Wilson, [12]according to the AP.
"First, Congress should establish an expert commission to study the means and methods of AI that can be used to exploit children specifically and to propose solutions to deter and address such exploitation," the document states.
[13]
"Second, after considering the expert commission's recommendations, Congress should act to deter and address child exploitation, such as by expanding existing restrictions on CSAM to explicitly cover AI-generated CSAM. This will ensure prosecutors have the tools they need to protect our children."
Addressing fake CSAM is tricky. Typical techniques that detect the illegal content relies on hashing known images that are being circulated online. It’s therefore difficult to identify new images, especially if they have been doctored using software. The Attorneys General, however, believe lawmakers must act because the technology will continue to evolve.
"We are engaged in a race against time to protect the children of our country from the dangers of AI. Indeed, the proverbial walls of the city have already been breached. Now is the time to act," the letter concludes. ®
Get our [14]Tech Resources
[1] https://regmedia.co.uk/2023/09/05/handout_ag_letter_csam.pdf
[2] https://www.theregister.com/2023/06/08/ai_deepfakes_sextortion_fbi/
[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.nassauda.org/CivicAlerts.aspx?AID=1512
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2023/06/08/ai_deepfakes_sextortion_fbi/
[9] https://www.theregister.com/2022/12/12/china_deep_synthesis_deepfake_regulation/
[10] https://www.theregister.com/2022/10/13/clientside_scanning_csam_anderson/
[11] https://www.theregister.com/2022/06/03/police_australia_ai/
[12] https://apnews.com/article/ai-child-pornography-attorneys-general-bc7f9384d469b061d603d6ba9748f38a
[13] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZPf5aNjLf8oFKjHHFkJneQAAAUY&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[14] https://whitepapers.theregister.com/
Some wrong assumptions are being made
Deepfakes and AI generated images are two completely different things. There should be no need to use live model child pornography to produce AI generated child pornography.
All the creators should need is images of pornography models who are of legal age but have slender builds, and non-pornographic images of children. They could then use generative AI technology to blend the two together.
I suspect that any law based on the premise that AI generated pornography depends on live model child pornography would fail in court if the defendant kept a good record of what data was used to train the model and could show that no live model child pornography was involved.
I suspect that the US will have to address the issue by either having a law that says that if it looks like child pornography, then it's child pornography even if no children were involved (which is what some countries do), or else deal with the issue of what constitutes a derivative work when AI is involved together with some privacy laws that have some real teeth and which can prevent images of children being used for unauthorized purposes. Then add on top of that a requirement for all image related AI models to keep a record of what training material they used.
In the end though, as AI oriented hardware acceleration becomes more mainstream I suspect that attempts to prevent illegal uses of it will become futile.