Straight Up Scene
- Amazon-supported AI firm Anthropic has found itself in hot water due to an "honest citation mistake" in a court filing, as claimed by the company.
- The mix-up involved an allegedly fictional academic article, which was cited to support the argument that Anthropic's AI model, Claude, reproduces copyrighted lyrics only under specific circumstances.
- Plaintiffs, including major publishers like Universal Music Group, dubbed the citation a "complete fabrication" and accused Anthropic of using its own AI to generate the erroneous source.
- In a San Jose hearing, the plaintiffs' lawyer Matt Oppenheim criticized the oversight, but believed the AI expert, Olivia Chen, didn't intentionally fabricate the citation.
- Anthropic's attorney, Sy Damle, argued that Chen's error was a simple mis-citation, not a fabrication, and that the issue was raised too late in the proceedings.
- The judge expressed concern over the matter, as the misuse of AI in generating citations could potentially mislead the court.
- Anthropic has been ordered to respond formally to the allegation, but has yet to comment on the issue to Decrypt.
AI AI, Go Away
AI company Anthropic allegedly uses AI-generated hallucination defense in a lawsuit over song lyrics ownership
The lawsuit against Anthropic was filed in October 2023, citing the unauthorized use of copyrighted lyrics to train the Claude model. The publishers demanded damages, disclosure of the training set, and the destruction of infringing content.
Anthropic responded in January 2024, denying any intentional copyright infringement. It labeled the reproduction of copyrighted lyrics as a "rare bug" and challenged the publishers to provide evidence of typical users encountering infringing content.
In August 2024, Anthropic faced another lawsuit from authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused the company of using pirated copies of their books to train the Claude model.
AI-itization of Copyright
The Anthropic lawsuit is part of an escalating legal battle between copyright holders and AI companies over the use of intellectual property in developing generative AI models.
OpenAI is facing lawsuits from comedian Sarah Silverman, the Authors Guild, and The New York Times, among others, for using copyrighted material without permission or licenses to train its GPT models. Meta is said to be implicated in similar suits, with plaintiffs arguing that its LLaMA models were trained on unlicensed literary works sourced from pirated datasets.
In March 2024, OpenAI and Google urged the Trump administration to loosen copyright restrictions for AI training, arguing these constraints hinder technological innovation[5]. In the UK, a bill allowing AI firms to use copyright-protected material without permission hit a snag this week, as the House of Lords backed an amendment requiring disclosure of materials used in AI models[6].
The General Buzz Newsletter
Extra Knowledge Nuggets
- The incorrect citation was generated using Anthropic's AI chatbot, Claude, which was not properly verified by the Anthropic legal team[1][2][3].
- The judge's concern over the misuse of AI in generating citations highlights the potential for AI to inadvertently create false or misleading information, which could have significant implications in legal proceedings[2].
- Anthropic's apology for the mistake mentions the steps they are taking to prevent such errors in the future, including improved human oversight and Verified Citations, a system designed to ensure the accuracy of AI-generated citations[3].
- History shows that similar incidents of AI-generated misinformation have occurred in the past, including the deepfake video of former President Barack Obama and the bogus academic papers generated by the AI researcher Varun Barramjee[7][8].
- The broader conflict between copyright holders and AI companies highlights the need for clear guidelines governing the use of copyrighted material in AI training. Until such guidelines are established, both parties will likely continue to face legal challenges and scrutiny[4].
- In light of the Anthropic lawsuit, questions arise about the ethical application of AI in the technology sector, particularly in the generation of citations.
- The controversy surrounding the alleged fabrication of a citation in Anthropic's court filing highlights the potential for AI to inadvertently create false or misleading information, which could have significant implications in technology and legal domains.
