Public consultation has begun on the new Artificial Intelligence Systems, a proposal by the Minister of Digitalisation that aligns with EU regulations stemming from the so-called AI Act.
The key part are the so-called regulatory sandboxes – controlled environments in which companies can test their innovative AI solutions, without the immediate risk of sanctions.
Why the regulatory sandboxes may not be sufficient
Are the sandboxes really the best way of supporting AI startups or are they just dead regulations that could potentially drain budgets.
Regulatory sandboxes – an opportunity or a limitation?
According to the assumptions of the law draft, regulatory sandboxes are meant to be safe environments where companies from the AI sector can test new artificial intelligence technologies under the supervision of a regulator. Theoretically, this model allows startups to avoid the risk of penalties for potential regulatory violations while providing them with the opportunity to experiment with innovations in controlled conditions.
However, in practice, sandboxes may encounter serious limitations, mainly related to technological infrastructure. In the AI sector, computational resources are key—powerful GPU cards, cloud computing, and advanced tools for training models.
Currently, there is a shortage of computational power in the market, especially graphic cards, which are essential for working on complex AI algorithms. Meanwhile, regulatory sandboxes, without external support, will not be able to offer computational resources at the level of leading providers such as Microsoft (Azure) or Amazon (AWS).
The problem with building infrastructure
Establishing regulatory sandboxes requires huge investments. Creating and maintaining infrastructure based on advanced computational technologies, such as GPUs, cloud servers, and security systems, is costly. The government would have to incur significant expenses to provide companies with a technological environment similar to the one offered by commercial cloud service providers. This raises the question of whether such an investment is cost-effective, especially considering a limited budget and other priorities.
For AI startups, which need scalable, advanced computational tools, regulatory sandboxes may prove insufficient. Companies using solutions from OpenAI, AWS, or Azure might prefer these platforms due to their scalability, reliability, and access to global technical resources. For the Polish regulator, creating a similarly advanced infrastructure seems practically impossible—at least without colossal financial outlays.
The Government Cloud
Startups and private enterprises, including those in the AI sector, will not have direct access to the Government Cloud, as according to regulations, these services are reserved exclusively for public administration entities and government-affiliated units.
An alternative approach: substantive support and access to computational resources
Instead of focusing on building their own regulatory sandbox infrastructure, the legislator could support companies better, through innovation acceleration and by providing access to affordable computational resources which are crucial for creating new models with suppliers already operating in the market. Here are a few ways to achieve this.
Subsidies and grants for cloud access
The government could provide companies with grants or subsidies for using existing platforms, without unnecessarily formalizing the process, for cloud solutions such as AWS, Azure, Google Cloud, or OpenAI API, etc. This would allow startups to train their AI models using the best available tools while minimizing costs for the state budget.
Example: A startup developing an image recognition system can receive funding to use computational resources on Amazon Web Services’ cloud, rather than being limited to the resources offered by regulatory sandboxes, which might not meet its requirements or restrict its possibilities.
Partnership with global providers
The government could collaborate with commercial tech giants such as Microsoft, Amazon, or Google to offer companies discounts or preferential terms for the use of their cloud services. Such partnerships could provide startups with access to advanced machine learning tools for a specified period of time, without burdening the state budget with infrastructure costs.
Example: A company working on AI chatbots could receive a discount for using Microsoft Azure, where it could train AI models on globally scalable infrastructure.
Accelerators and scientific support
The government could invest in accelerator programs which, in addition to financial support, would also offer training and mentoring in AI, not only helping startups build technology but also helping them connect with universities and supporting them in complying with regulatory requirements. The government could also, within the sandboxes, support collaboration with universities and other technological entities that can provide knowledge and recommendations assistance for companies in bringing safe products to the market more quickly.
AI ACT. Are sandboxes actually going to help startups and consumers?
In model currently proposed, no entrepreneur is going to voluntarily use sandboxes, unless they don’t have another choice. Instead, it would be better to offer legal assistance, consultations and resources for testing products in terms of compliance with established regulations, and to provide computational resources through commercial suppliers.