A rights-free zone? Blank national security exemption in AI legislation

It is increasingly recognised that the use of Artificial Intelligence (AI), while promising benefits in many areas, can also pose serious threats to fundamental rights. To counter these threats and to protect those rights, binding rules on the use of AI systems are being proposed at both European Union (EU) and Council of Europe (CoE) level:

  1. in the EU, the AI Act (AIA) is already going through the legislative process;
  2. in the CoE, a so-called “zero draft” of a Convention on AI, Human Rights, Democracy and the Rule of Law has been circulated.

However, European legislators are seeking to exclude from the application of the proposed instruments the use of AI in the contexts in which they arguably pose the greatest threats to fundamental rights: national security, defence and transnational law enforcement.

In a legal opinion prepared for the European Centre for Not-for-profit Law, I argue that AI systems that are unexplainable/ unchallengeable present an unacceptable risk and should be prohibited in both the AIA and the CoE convention. Specifically:

  • There cannot be a blanket national security exemption in the AIA, as any exemptions must be proportionate and not undermine EU law, e.g., the EU internal market or data protection legislation.
  • Europol, Frontex and other EU agencies, are not exempt from the obligations of the Charter of Fundamental Rights of the EU and should not be exempt from the AIA.
  • AI systems developed for military or dual use should not be exempt from the AIA.
  • EU foreign and defense policies/ instruments should be reviewed to reflect a prohibition of unexplainable/ unchallengeable AI systems.
  • The design, development and use of AI systems for national defense should not be explicitly excluded from the CoE convention.