On 2 August the deadline for having a national AI sandbox expired, and inside those sandboxes data collected for other purposes may be processed

What a regulatory sandbox is

It is a controlled environment in which an AI system is developed, trained, tested and validated for a limited time, under the supervision of the competent authority and according to an agreed plan. It is not a free zone: the rules continue to apply, but the conversation with the authority happens before placing on the market rather than after, and a participant comes out with documentation the authority has already looked at.

Paragraph 3 of the same article adds something that is often missed: the European Data Protection Supervisor may also establish a sandbox for Union institutions, bodies, offices and agencies, exercising the roles and tasks of national competent authorities.

Why this concerns anyone working on personal data

Because of Article 59. In the sandbox, personal data lawfully collected for other purposes may be processed solely for the purpose of developing, training and testing certain AI systems. That is an express provision for reuse, and EU law does not contain many. But it holds only if all the listed conditions are met, not some of them.

The system must serve a substantial public interest, in one of these areas

• Public safety and public health, including disease detection, diagnosis prevention, control and treatment and improvement of health care systems.
• A high level of protection and improvement of the quality of the environment, protection of biodiversity, protection against pollution, green transition measures, climate change mitigation and adaptation measures.
• Energy sustainability.
• Safety and resilience of transport systems and mobility, critical infrastructure and networks.
• Efficiency and quality of public administration and public services.

And then all the other conditions have to hold too

• The data must be necessary for complying with the requirements of Chapter III, Section 2, and those requirements must not be effectively satisfiable with anonymised, synthetic or other non-personal data.
• There must be effective monitoring mechanisms to identify high risks to the rights and freedoms of data subjects, as referred to in Article 35 GDPR, and response mechanisms to mitigate them promptly and, where necessary, stop the processing.
• The data must sit in a functionally separate, isolated and protected processing environment under the control of the prospective provider, accessible only to authorised persons.
• The originally collected data may be further shared only in accordance with Union data protection law, and personal data created in the sandbox cannot leave it.
• The processing must neither lead to measures or decisions affecting the data subjects nor affect the exercise of their rights.

What to ask now

• Whether your own State's sandbox is operational and on what terms it can be entered: the deadline is in the Regulation, implementation varies from State to State and has to be checked with your own authority.
• Whether the client is considering an AI project in one of the five areas, because in that case the sandbox is not a nicety but the most defensible route.
• Whether anyone in the organisation has already written that they will train on existing data citing the AI Act as the basis: this is the moment to reread Article 59 line by line.