Carding Forum
Professional
- Messages
- 2,788
- Reaction score
- 1,176
- Points
- 113
A group of scientists and hackers is seeking permission to violate the terms of use of AI companies for the sake of good-faith research that reveals biases, inaccuracies and training data, without the risk of prosecution.
The US government is considering making an exception to the copyright protection legislation, which will allow circumventing protective measures in AI systems. This will provide an opportunity to explore the workings of AI, identify bias, discrimination, and inaccurate conclusions, and learn more about training data. The exception will allow you to conduct bona fide security research and academic research, even if it requires bypassing security systems.
The US Department of Justice supports the proposal, arguing that good-faith research can reveal data leaks and systems whose conclusions are insecure or inaccurate. This is especially important when AI is used for meaningful purposes, where mistakes can cause serious harm.
Most of what is known about the work of closed AI tools such as ChatGPT and Midjourney became known thanks to researchers, journalists and users who deliberately try to deceive systems in order to learn more about the data on which neural networks are trained, their biases and weaknesses.
However, such research often violates the terms of use. For example, the OpenAI terms of use prohibit attempts to reverse engineer or discover the source code of services, as well as circumvent security measures.
MIT researchers promoting the proposal noted that there are many concerns about AI models, their structure, bias, and use for discrimination. However, researchers often get their accounts blocked for doing good research or fearing legal consequences. Such conditions hinder research, and companies are not always transparent about their compliance.
The exception will apply to Section 1201 of the Digital Millennium Copyright Act (DMCA). Other exceptions to this section already allow hacking devices for repair and protect security researchers who try to find bugs and vulnerabilities, and in some cases protect those who try to archive or save certain types of content.
There are many examples of academic articles, investigative journalism, and research projects that require hacking or cheating AI tools to detect training data, bias, or unreliability.
The authors of the proposal published an analysis that mentions the previous terms of use of Midjourney, which threaten to recover damages for intellectual property infringement. The researchers argue that AI companies have begun to use their terms to constrain analysis.
Many studies focus on trying to get AI to reproduce copyrighted works in order to prove that the models are trained on such materials. This tactic was used by the music industry to prove that the Suno and Udio AI tools were trained on copyrighted music, which became the centerpiece of a lawsuit against the companies. You can imagine a scenario where researchers use AI to reproduce protected works, which leads to negative consequences for the AI company that accuses the researcher of violating the terms of use.
The Copyright Office of Harley Geiger of the Hacking Policy Council noted that the exception is critical for identifying and correcting algorithm errors. The lack of clear legal protections under section 1201 of the DMCA negatively affects such research. The exception will not stop companies from trying to prevent such research, but it will legally protect researchers who violate the terms of use in order to conduct an analysis.
At a hearing this spring, Morgan Reed of the App Association, which represents many AI companies, argued that researchers should get prior consent from AI companies to conduct the analysis. Without notification, the specialist is actually considered a potential malicious hacker. Reed stressed that researchers want to get protection from liability after the fact.
The DVD Copy Control Association, which represents major film studios and is one of the pioneers of DRM, also opposes the proposed exception.
The US government is considering making an exception to the copyright protection legislation, which will allow circumventing protective measures in AI systems. This will provide an opportunity to explore the workings of AI, identify bias, discrimination, and inaccurate conclusions, and learn more about training data. The exception will allow you to conduct bona fide security research and academic research, even if it requires bypassing security systems.
The US Department of Justice supports the proposal, arguing that good-faith research can reveal data leaks and systems whose conclusions are insecure or inaccurate. This is especially important when AI is used for meaningful purposes, where mistakes can cause serious harm.
Most of what is known about the work of closed AI tools such as ChatGPT and Midjourney became known thanks to researchers, journalists and users who deliberately try to deceive systems in order to learn more about the data on which neural networks are trained, their biases and weaknesses.
However, such research often violates the terms of use. For example, the OpenAI terms of use prohibit attempts to reverse engineer or discover the source code of services, as well as circumvent security measures.
MIT researchers promoting the proposal noted that there are many concerns about AI models, their structure, bias, and use for discrimination. However, researchers often get their accounts blocked for doing good research or fearing legal consequences. Such conditions hinder research, and companies are not always transparent about their compliance.
The exception will apply to Section 1201 of the Digital Millennium Copyright Act (DMCA). Other exceptions to this section already allow hacking devices for repair and protect security researchers who try to find bugs and vulnerabilities, and in some cases protect those who try to archive or save certain types of content.
There are many examples of academic articles, investigative journalism, and research projects that require hacking or cheating AI tools to detect training data, bias, or unreliability.
The authors of the proposal published an analysis that mentions the previous terms of use of Midjourney, which threaten to recover damages for intellectual property infringement. The researchers argue that AI companies have begun to use their terms to constrain analysis.
Many studies focus on trying to get AI to reproduce copyrighted works in order to prove that the models are trained on such materials. This tactic was used by the music industry to prove that the Suno and Udio AI tools were trained on copyrighted music, which became the centerpiece of a lawsuit against the companies. You can imagine a scenario where researchers use AI to reproduce protected works, which leads to negative consequences for the AI company that accuses the researcher of violating the terms of use.
The Copyright Office of Harley Geiger of the Hacking Policy Council noted that the exception is critical for identifying and correcting algorithm errors. The lack of clear legal protections under section 1201 of the DMCA negatively affects such research. The exception will not stop companies from trying to prevent such research, but it will legally protect researchers who violate the terms of use in order to conduct an analysis.
At a hearing this spring, Morgan Reed of the App Association, which represents many AI companies, argued that researchers should get prior consent from AI companies to conduct the analysis. Without notification, the specialist is actually considered a potential malicious hacker. Reed stressed that researchers want to get protection from liability after the fact.
The DVD Copy Control Association, which represents major film studios and is one of the pioneers of DRM, also opposes the proposed exception.
