These days, it seems like the latest news on new generative AI technology is everywhere.
Generative AI represents a true paradigm shift in our ability to create content and ideas, with the potential to have significant implications for the way we work and live.
In the child safety ecosystem, generative AI is also changing how child sexual abuse happens online and how we can combat it.
As technology companies and organizations race to keep up with new advances in generative AI, Thorn is working hard to advocate for children as it develops – just like we do with all other technologies as they emerge.
The growing prevalence and use of this technology already has many implications for child safety – most of which the child safety ecosystem as a whole is only just beginning to understand and analyze.
However, one thing we TO DO know at this point is that now is the time for security by designand AI companies must lead the way to ensure child protection as generative AI technology is developed.
now is our chance
While all we know indicates that this new wave of generative AI technology could pose serious threats to children, we see a silver lining: we have a unique opportunity to act. NOW put child safety at the center of this technology as it emerges.
In fact, we believe we are one timely time to step into that space – and now is the time for safety by design.
Security by design encourages thoughtful development: Rather than modernizing protections after a problem has occurred, technology companies should be thinking about how to minimize threats and damage throughout the development process. For generative AI, this concept must be extended to the entire machine learning (ML)/AI lifecycle: development, deployment and maintenance. Each of these parts of the process includes opportunities to prioritize child safety.
Taking into consideration developmentsome tactical actions that can be taken include:
- Remove harmful content from training data, for example by hashing and matching data against known CSAM hash sets, or using classifiers and manual review.
- Engage in “red teaming” sessions (the practice of stress testing systems – physical or digital – to find flaws, weaknesses, gaps and edge cases) to pressure test particular themes and content, for example, which prompts produce AI-generated child sexual abuse material (CSAM).
- Incorporate technical barriers to the production of harmful content, for example, biasing the model against the production of child nudity or sexual content involving children.
- Be transparent with training packages (especially in an open source setting) so collaborators can independently audit/evaluate content for harmful content.
Taking into consideration deployment:
- For cloud-based systems, integrate harmful content detection into the inputs and outputs of that system, for example, detecting prompts intended to produce AIG-CSAMs and detecting AIG-CSAMs that may have been produced.
- For open source systems, assess the platforms you allow to share your technology, for example, determine if these platforms knowingly host models that generate harmful content.
- For platforms that share models developed by other organizations and people, evaluate which models you allow to host on your platform, for example, only host models that have been developed with child safety in mind.
- In any case, look for content provenance solutions that occur as part of development rather than as an optional post-processing step, such as forming a watermark in the model decoder itself or publishing ML/AI solutions that can reliably predict the synthetic nature of content.
Taking into consideration interview:
- As new models are developed and deployed with security-by-design principles, remove access to historical models.
- Proactively ensure synthetic content detection solutions perform well on content generated by new models.
- Actively engage with special interest groups to understand how your models are being misused.
- For cloud-based systems, include clear pathways for reporting violations to the proper authority.
- Share known AIG-CSAM hashes and known entries that produce harmful content discovered during this process with the child safety ecosystem.
How Thorn Helps
Navigating the landscape of emerging technologies is nothing new for Thorn; for more than a decade, we have stayed ahead of the curve, attentive to the implications that new technologies can have for children and used technological solutions to solve technological problems.
We are unique in this space, connecting knowledge and fostering collaborations across the child safety ecosystem. At the same time, our dedicated team of data scientists are solely focused on understanding the landscape and developing solutions to the unique security challenges we face, including challenges presented by new technologies such as generative AI.
One of the ways we facilitate this is through our Safer product, a solution designed to help companies detect, identify and report CSAM at scale.
With resources like Safer, as well as our consulting expertise, strong child safety relationships, and other targeted offerings – such as red team sessions – for platform safety, Thorn is uniquely positioned to work with the generative AI industry to bring child safety to the forefront of innovation.
We are also encouraged that many key players in the generative AI space are willing to work with Thorn – as well as other players in the ecosystem, including the National Center for Missing and Exploited Children (NCMEC), THE Technology Coalitionand others – to protect children and build responsibly.
For example, Open AI has integrated Safer into its DALL·E 2 generative AI web application. This collaboration illustrates how we can proactively respond to potential threats, integrating safeguards into the fundamental structure of the technology.
As we continue to monitor and adapt to the changing AI landscape, we are always ready to help others do the same with our own AI solutions and technology. We believe in the power of collective action. By building strong partnerships, sharing knowledge and leveraging relationships, we can ensure a safer technological future for children. And it all starts with us, together, prioritizing safety by design.
