The leading cloud computing firms globally, which have introduced innovative artificial intelligence solutions to their clientele, are offering only a limited set of protections against potential copyright litigations related to these systems.
In a massive bet on relational AI—technology capable of swiftly generating human-like text, images, and scripts—Amazon, Microsoft, and Google are in competition to deliver new offerings such as virtual assistants and chatbots.
These AI models are “trained” using data sourced from the internet, encompassing images and text. Concerns have escalated regarding third parties utilizing AI tools that are based on copyrighted data, particularly impacting media entities and image repositories.
While the top three cloud computing giants have committed to safeguarding businesses and consumers from intellectual property disputes, these legal safeguards are specifically applicable to designs produced by or under the auspices of Google, Amazon, and Microsoft, as per an analysis of the compensation clauses outlined by these companies in the cloud computing sector.
These stipulations represent a prudent strategic move. Matthew Sag, a law professor at Emory University, remarked, “… and give people the confidence to utilize this technology without apprehension.”
However, Brenda Leong, a partner at Luminos Law, emphasized the importance for enterprises to grasp the precise definitions and scope of these provisions.
Microsoft, Amazon, and Google opted not to provide comments on this matter.
Despite the availability of these tools on the cloud providers’ platforms, the protections extended to buyers do not encompass the utilization of third-party models, such as those developed by AI startup Anthropic, which has investments from Amazon and Google.
For instance, Amazon’s coverage only includes content generated by its proprietary models like Titan and various AI software.
Similarly, Microsoft offers protection solely for tools created by OpenAI, the startup it collaborates with through a multimillion-dollar alliance, and those operating on its in-house models.
An intellectual property attorney addressing these concerns mentioned, “People required these safeguards because they were acutely aware of the legal risks involved.”
Simultaneously, the three cloud service providers are fortifying their tools with safety mechanisms to filter out potentially objectionable content. The attorney noted that while the tech companies were increasingly confident that breach incidents would be minimal, they were cautious about providing unrestricted protection.
Clients of Microsoft, Amazon, and Alphabet might consider negotiating more tailored indemnities in contracts to suit their specific requirements, although this practice is not yet widespread, according to sources close to these cloud firms.
Notably, legal battles over conceptual AI test cases initiated by prominent figures like authors and comedian Sarah Silverman are being contested by entities such as OpenAI and Meta. These disputes primarily revolve around allegations that the companies training the models infringed copyright laws.
To mitigate concerns about potential litigations, additional protective measures have been extended to individuals, according to Angela Dunning, a partner at Cleary Gottlieb law firm. This is due to the protracted resolution timelines of such cases, leading to a period of uncertainty.
However, unlike Microsoft, Google’s compensation does not encompass models fine-tuned by customers using internal organizational data, a practice that enables businesses to enhance basic models for more precise outcomes.
While Amazon covers Titan variants that have been customized in this manner, the defense is revoked if the alleged copyright infringement is linked to the fine-tuning process.
Challenges in prevailing legal claims against users rather than creators of relational AI tools may arise.
US Judge William Orrick highlighted that a key challenge was proving that every image generated by these tools was derived from “copyrighted training images,” dismissing a portion of a claim filed by three artists against AI firms Stability AI, DeviantArt, and Midjourney last year.
Orrick emphasized that the AI-generated content must be “substantially related” to the copyrighted material to constitute infringement.
