Industry Statement on the Risks of GPAI Regulation for Startups in the EU

Currently, a number of regulations in the digital space are either being enacted or enforced, which is forcing tech companies to find ways to comply with this tsunami of EU regulations. Existing smaller actors, who have inherently less resources than more established players, are struggling to digest and comply with regulation. This regulatory hostility disincentivises startup creation. This is especially true in AI, where thousands of startups are innovating. In addition to the overburdening responsibilities that developers and deployers of so-called “high-risk AI systems” would bear based on the Commission’s proposed text, many policymakers and stakeholders are calling for the final version of the AI Act to also cover general purpose AI systems (GPAIs) as well as foundation models.

Allied for Startups and its Members believe that regulating GPAIs and foundation models will have detrimental effects on the startup ecosystem for three reasons:

First, it fuels legal ambiguity at a time when smaller players, such as startups, need regulatory stability more than ever. Contradicting the AI Act’s risk-based and technology-neutral approach by introducing vertical provisions will create an uncertain environment where startups will not be able to identify whether an AI is risky based on its intended purpose. Since a GPAI doesn’t have a specific purpose by definition, introducing provisions covering GPAI (as well as foundation models and generative AI systems, as proposed by the European Parliament), would run counter to the regulation’s main objective and be detrimental to the startup ecosystem.

Second, burdensome GPAI obligations affect nascent startups and their ability to scale. Startups are very innovative and scalable businesses that need to use all the tools at their disposal to grow and prosper, including GPAI systems. A survey from December 2022 shows that 45% of the startups consider their AI system to be a GPAI, making them subject to these hypothetical obligations, which would essentially create entry barriers to the GPAI market and disincentivise the development of such systems for startups to use.

Third, this regulatory avalanche will create multiple overlaps with existing laws. Introducing copyright-related obligations for providers of generative AI systems is unworkable in practice (providers would be required to provide a ‘detailed summary’ of all the data used to train their generative AI system) and touches upon an area that the Copyright Directive already covers, thus duplicating regulation.

Consolidating the current policy framework by enforcing existing laws and assessing the effectiveness of said laws before even considering enacting new ones, provides startups with the regulatory stability for a healthy and thriving startup ecosystem. This is especially true in AI, any regulatory approach of GPAI should respect the AI Act’s risk-based approach, take into account the financial and technical realities of startups, and avoid hampering innovative generative AI systems.