ANI Takes Legal Action Against OpenAI Over Copyright Infringement and Content Usage

ANI Seeks Court Order to Stop OpenAI from Using Its Content

Disclaimer: This content generated by AI & may have errors or hallucinations. Edit before use. Read our Terms of use News agency Asian News International (ANI) asked the Delhi High Court to prohibit OpenAI from using its data for future training while requesting an interim injunction against the AI company. ANI's lawyer, Sidhant Kumar, argued that OpenAI's previous commitment to not use ANI's content to train its AI model was insufficient due to the nature of the agency's business. Kumar pointed out that ANI produces news articles, photographs, and videos through its own network of journalists and licenses them to other news platforms for syndication. He explained that this arrangement allows other news platforms to republish ANI's content as is or use parts of it with their own inputs. However, they are not permitted to syndicate it further. Kumar emphasized that OpenAI could easily access ANI's content from these news platforms. Therefore, ANI requested the court to ensure that OpenAI refrains from collecting ANI's licensed content not just from the agency's own website but also from its subscribers. ANI also accused OpenAI of causing "irreparable harm" to its business by diluting the market for its licensed content. Kumar highlighted that OpenAI had entered into content licensing agreements with other news organizations. "Firstly, you acknowledge that this content does have copyright protection, and therefore you have entered into these agreements. On the other hand, you are restricting my market because you are, if not a news agency, distributing other parties' news. Therefore, you are restricting the market for licensed content as far as news is concerned," said Kumar. ANI's lawyer referred to the statements made by the Amicus Curiae in a previous hearing. "The suit, as it's framed today, is based on four different stages of infringement, which are independent and self-sustaining on their own," he said. The first three stages occurred during the training phase of the model, while the last involved ChatGPT reproducing ANI's licensed content. Kumar stated that even if the court did not agree that ChatGPT's responses constituted copyright infringement, ANI would still have grounds for relief if the training process itself violated copyright. He pointed out that Section 14 of the Copyright Act grants the copyright owner the exclusive right to reproduce literary work 'in any material form, including the storing of it in any medium by electronic means.' "So that I have the exclusive right to do. And any breach of an exclusive right is infringement," he said. He also pointed out that the Act granted him the exclusive right to communicate the work to the public, adapt the work and translate it. Kumar also cited a previous judgment by Justice Endlaw, which stated that Section 52 -- providing exceptions to copyright infringement -- was all-encompassing. This meant that there was no copyright exception other than what the section contained. ANI also rejected a point made by Amicus Curiae Prof. Scaria in the previous hearing, who questioned whether ANI even held the copyright to its works, as it might not have an employer-employee relationship with the people who produced its content. Kumar countered that ANI had signed contracts of service with the creators, affirming ANI as the first owner of the copyright. The news agency had sued OpenAI, the creator of ChatGPT, over copyright infringement last year. It argued that the AI company had used ANI's copyrighted content to train its Large Language Models (LLMs) without authorization. ANI also claimed that ChatGPT was able to reproduce its content verbatim in response to user prompts and, in some cases, had even attributed events that never occurred to the agency's reportage.

Usage Of Content By OpenAI May Cause Unfair Competition: ANI To Delhi HC Usage Of Content By OpenAI May Cause Unfair Competition: ANI To Delhi HC

Advocate Siddhant Kumar, who is representing ANI in its suit against OpenAI reportedly told Delhi HC that ChatGPT first used its content to train a language model, stored that data and then made it publicly accessible Amid the ongoing legal battle between Indian media outlets and tech giant OpenAI over alleged copyright infringement, news agency ANI reportedly told the Delhi High Court (HC) that the usage of its content by the ChatGPT maker causes dilution of its market, thus leading to unfair competition. The Delhi HC was hearing a petition filed by ANI last year, accusing OpenAI of using its news content without authorisation. This comes amid growing concerns of news publishers in multiple jurisdictions, including India and the US, over the use of their copyrighted material by AI companies like OpenAI to train their foundational models, without licence, or permission or payment. During a court hearing on Monday (March 11), advocate Siddhant Kumar, who is representing ANI in its suit against OpenAI, submitted that ChatGPT first used its content to train a language model, stored that data and then made it publicly accessible, Business Standard reported. Amicus curiae Adarsh Ramanujan said that while ANI's data was not used directly and its content was published behind a paywall, subscribers with varying levels of subscription restrictions could republish the content. In court parlance, an amicus curiae is someone who is not a party to the case but can give information or offer advice to courts. He further said that mere availability of content in the public domain does not nullify copyright protections and added that the Indian law does not explicitly recognise exceptions outside the fair use doctrine, the report said. The Delhi HC has scheduled the next hearing in the case on March 20. It must be noted that several Indian digital news publishers, including the likes of NDTV, Network18, The Indian Express and Hindustan Times, have joined ANI's legal battle against OpenAI, seeking to prevent the company from accessing their proprietary content. The Sam Altman-led AI giant, however, has argued that it is not obligated to enter into partnerships with these media outlets to use their content and urged the HC to dismiss infringement claims filed against it. Major music labels such as T-Series, Saregarama and Sony have also expressed their willingness to join the ongoing copyright lawsuit against the ChatGPT developer in the Delhi HC. Consequently, the HC asked OpenAI to file a response regarding the application filed by the Indian Music Industry (IMI) seeking to intervene in the suit.

ANI vs OpenAI: Is Storing Copyrighted Content An Infringement?

Disclaimer: This content generated by AI & may have errors or hallucinations. Edit before use. Read our Terms of use Storage of copyrighted material amounts to infringement, told amicus curiae Adarsh Ramanujan to the Delhi High Court on March 10. The court had appointed Ramanujan as an expert to guide the court in the ongoing copyright infringement lawsuit between news agency ANI and OpenAI. The amicus curiae argued that collecting public data necessarily involved storing it, which qualified as "reproduction" under Section 14(a)(i) of the Copyright Act, 1957, making it an infringement of copyright under Section 51. "There is infringement because it's strict liability. Intent and purpose do not matter," he said. Ramanujan also added that unlike other jurisdictions, Indian copyright law did not have an exemption for Text and Data Mining (TDM). He was referring to the fact that OpenAI collected vast quantities of data from the internet to train its Large Language Models (LLMs). ANI had objected to this practice, arguing that it amounted to copyright infringement. The news agency had also claimed that ChatGPT, OpenAI's AI-powered chatbot, was capable of reproducing verbatim copies of ANI's content in response to user queries and had even attributed events that never occurred to ANI's reporting. Ramanujan elaborated on the development process of Large Language Models (LLMs), splitting them into three parts: Ramanujan argued that steps one and three could involve the infringement of copyright while step two could not. He alluded to the submissions of Arul George Scaria, the other amicus curiae in the lawsuit, who had pointed out that the training process involved only a non-expressive use of copyrighted material, and thus did not qualify as copyright infringement. Ramanujan agreed with this assertion, but limited it to only the second step of the process. "But preceding this non-expressive use and post this non-expressive use, there is still expressive use, that is infringement," he explained. "The fact that something is published and freely available does not mean copyright goes away," said Ramanujan. He was referring to the fact that ANI syndicated much of its content to other news platforms, which may publish them widely without paywalls. OpenAI used crawlers to scrape data from the internet, which it then used to train its LLMs. As such, it was able to access ANI's content without having to scrape the agency's own website. Ramanujan also addressed the Robots Exclusion Protocol, a widely accepted industry standard by which website owners could direct third-party crawlers about which web pages they are allowed to access. However, he pointed out that copyright law did not place the onus on copyright owners to safeguard their property. "An infringer cannot contend that the plaintiff did not safeguard the copyrighted work to prevent infringement," he said. Ramanujan further argued that there was no regulation or law requiring crawlers to comply with the protocol. The amicus curiae also stated that just because OpenAI had agreed to stop accessing ANI's website in October last year, it was not a defence against past infringement. Ramanujan argued that during the process of tokenising and vectorising the collected data, there was no infringement of copyright. During this process, each expression was broken down and assigned a numerical identifier, which in turn was represented by another set of numbers. These numbers gave the program a mathematical idea of the relationship between different words or parts of words. "This is absolute non-expressive use because it does not reproduce the original expression in any way. So, at this stage, there is no infringement at all in my submission," said Ramanujan. He also described the process of LLM training, stating that a training loop existed where the LLM must iteratively predict the next answer while checking in self-supervised mode if its prediction is correct or close to what it should predict. He explained that if the prediction is incorrect, the LLM changes its parameters and repeats until it approaches the original expression. Ramanujan pointed out that what remains undisputed, even by the defendant, is the iterative nature of training where model parameters are updated with each iteration to improve predictions. He argued that in order for this process to work, the LLM would have to compare its answer with the original training data. This was important, as it would imply a reproduction of the original copyrighted material. "At the training stage also...there has to be necessarily a reiterative production of the original data because without that reiterative reproduction, you cannot compare, and if you do not compare, you cannot train" he explained. Ramanujan bolstered his claim by citing the testimony of an AI professor who had made similar statements in front of a US House of Representatives Subcommittee. Ramanujan addressed ChatGPT's tendency to occasionally regurgitate verbatim expressions of copyrighted material, despite being a probabilistic prediction machine. He provided examples of ChatGPT reproducing significant parts of the Harry Potter and Dr. Seuss books, even though users had not explicitly asked for those. He stated that there were a number of theories to explain this behaviour and that OpenAI would ultimately have to provide an explanation. Ramanujan accepted that OpenAI had placed a manual block to prevent ChatGPT from reproducing copyrighted material. "If you are developing a tool that is capable of doing this as a defendant, isn't the onus on you to prevent this before, rather than wait for somebody to point out infringement and then correct it?" he asked. Ramanujan stated that ANI would have to establish the following before the court: He stated that the first point was critical as ANI was in the business of news, which copyright law did not always cover. However, Ramanujan argued that there was no absolute rule that stated the above, as there may be expressive content in news items as well. If ANI is able to establish that ChatGPT's output contains expressive elements of news articles, it would count as "communication to the public," which is an exclusive right granted by Section 14 of the Act to the copyright owner. Therefore, OpenAI would stand in violation of copyright. Ramanujan argued that Section 52(1) of the Copyright Act, which granted exceptions to copyright infringement, was in fact a narrow list that only granted very specific exemptions. He also added that each exception only applied to certain exclusive rights from Section 14 and not to others. He argued that OpenAI would have to show that its activities came under the specific fair dealing exemptions from Section 52(1)(a), which covers private use, criticism and the reporting of current affairs. "If the defendant is unable to show that what it is doing is within the parameters of (i),(ii),(iii) [under Section 52(1)(a)] there cannot be a fair dealing exception applied to it," said Ramanujan.