(Photo by Jonathan Raa/NurPhoto via AP)
On August 31, the European Commission made headlines by officially designating OpenAI’s ChatGPT as a Very Large Online Search Engine (VLOSE). This significant decision places the widely adopted chatbot under the most stringent obligations set forth by the Digital Services Act (DSA). The move symbolizes a substantial shift in digital regulation, particularly concerning AI technologies. This decision, years in the making, is poised to have broad-ranging implications not only for products stemming from frontier labs but also for the regulatory landscape in the EU as these developments unfold.
### Understanding the VLOSE Designation
The VLOSE designation confirms that the DSA is technologically neutral and adaptable to emerging services that may not have existed at its inception. Article 3(j) of the DSA defines a search engine as a service allowing users to search “in principle, all websites.” As ChatGPT prepares to roll out a version with web access, arguably aligning perfectly with the definition, it becomes evident why the Commission sought to classify it as such.
ChatGPT is no longer merely a conversation piece; it incorporates a fully integrated search function that can even be set as the default for many browsers. This versatility resonates with the Commission’s findings and aligns with the transparency obligations laid out for online search engines, set to remain applicable as early as 2024. Under this new designation, ChatGPT will be subject to various due diligence requirements, including risk assessments, mitigation measures, and periodic independent audits, reinforcing accountability within AI systems.
### The Hybrid Nature of ChatGPT
More Read
Despite the impressive statistics of 159.1 million search users, it’s important to note that this figure represents only a fraction of ChatGPT’s broader user base within the EU. The interface, allowing users to engage with the AI model, doesn’t always perform web searches, making it a “hybrid service.” This classification complicates the regulatory landscape, as the designation does not distinctly separate ChatGPT’s functionalities, consistent with how past VLOP designations have worked—taking Snapchat as an example, designated for its social networking component rather than its messaging features.
The future of how the chat function will be categorized under Article 34(1) remains uncertain. OpenAI is now compelled to assess risks associated with the design or functionality of its service and related systems. There’s a solid argument to be made that the chat functionality could indeed qualify as a “related system,” warranting specific scrutiny and investigation.
### Risks and Responsibilities
Beyond mere legal obligations, the designation under the DSA paves the way for a more profound examination of ChatGPT’s inner workings. The genesis of a general-purpose AI model, from its training data to output ranking, has direct implications on the relevance, inclusivity, and reliability of the information provided to users. According to Recital 84 of the DSA, algorithmic amplification and curation are systemic risks that align with the mechanisms of how ChatGPT chooses and ranks sources, thus impacting democratic discourse as well as public well-being.
Notably, media pluralism has been flagged as a systemic risk distinctively categorized under Article 34(1)(b). This calls for a thorough risk assessment focusing on how ChatGPT’s citation practices influence publisher visibility. Such insights are not limited to OpenAI but resonate with broader industry concerns. Organizations like AlgorithmWatch are already leveraging Article 40(4) access to assess if AI-generated summaries are undermining media visibility, showcasing the urgency of this dialogue.
### Enhancing Accountability through Data Access
The VLOSE designation offers OpenAI greater transparency obligation, falling under the Article 40 data access framework. This is advantageous as it grants external researchers the capability to probe deeper into potential systemic risks associated with ChatGPT’s functionalities. Data requests can now be made to investigate how well OpenAI’s measures mitigate risks and impacts related to democratic processes, public security, and other critical societal dimensions. Notably, non-vetted researchers, including those affiliated with nonprofit organizations, are entitled to access public data available on ChatGPT’s interface.
### The Timing of Regulation and Advertising Changes
Interestingly, this regulatory decision comes shortly after OpenAI’s announcement that advertisers will be able to place ads within ChatGPT across Europe. This development means that OpenAI must now incorporate its advertising model into the ongoing risk assessment process and publish an ad repository under Article 39 of the DSA. As these obligations come into effect in January 2027, the stage is set for a more regulated landscape for generative AI and how it interacts with advertisers and consumers alike.
### Future Implications Under the DMA
Even though generative AI is not explicitly treated as a core service by the Digital Markets Act (DMA), the regulation of online search engines places ChatGPT in a unique position. With its recent designation as a VLOSE under the DSA framework, the possibility of a gatekeeper designation under the DMA now appears much more plausible. This future designation would impose an even more stringent set of qualitative and quantitative criteria on OpenAI, intensifying the scrutiny of its practices in the near term.
As the European Commission’s oversight progresses, how these complex issues will unfold remains an intriguing subject of consideration. The ongoing development of regulatory measures will undoubtedly shape AI technologies and their societal impacts in profound ways.
Inspired by: Source
