Council of Europe creates rules for Artificial Intelligence

Algorithmic systems – often referred to by the buzzword Artificial Intelligence (AI) – increasingly pervade our daily lives. They are used to detect social benefits fraud, to surveil people at the workplace, or to predict parolees’ risk of reoffending. Often, these systems do not only rest on shaky scientific grounds but can be used in ways that infringe people’s basic rights – like those to non-discrimination, freedom of expression, privacy, or access to justice –, can undermine foundational democratic principles, and through their non-transparent nature and the lack of accountability mechanisms can be in tension with the rule of law. Against this background and in light of its mandate, the Council of Europe had recognized the need for states to govern the development and use of AI systems: To this end, its member states – among which Switzerland – and selected other interested states – like the US or Japan – had been negotiating a Convention in AI. Civil society organizations like AlgorithmWatch, experts, and companies also participated in the negotiations as observers.

The Council of Europe is an international organization founded in 1949 with the task to uphold human rights, democracy, and the rule of law in Europe. It currently has 46 Member States (27 of which are also members of the European Union) and is based in Strasbourg. It is not to be confused with the European Council and the Council of the EU, which are both EU bodies – in contrast to the Council of Europe. In 1950, it drafted the European Convention on Human Rights (ECHR), the ratification of which still is a condition for new members to join. The Council of Europe hosts the European Court of Human Rights, which oversees the implementation of the ECHR, and promotes human rights through a range of additional measures, including international conventions, such as the Convention on Cybercrime or Convention 108 on data protection.

What would this mean? How will the outcome of the Council of Europe’s work protect our rights, our democracy, and principles of rule of law?

The Council of Europe’s plan created an international "Framework Convention," which is an international treaty that will become legally binding under international law upon states that sign on to it. States would thus be free to sign or not to – but if they do, they legally commit themselves to comply with it. Signatures will be open to non-member states, which is why states like the US or Israel also participated in the negotiations.

The Convention contains a range of obligations for states to make sure that human rights are respected in the development and use of AI systems. That said, it is often referring to general principles and leaves many aspects open to interpretation and implementation. Signatory states will be required to implement the provisions at the domestic level, i.e., through introducing domestic measures and laws. The requirements are generally applicable to all sectors – but they are not as stringent for private companies as they are for public authorities: States have more leeway to decide on which measures they implement for private companies, and these measures do not have to include binding laws. And public authorities are exempt from the Convention in the field of "national security," to which the Convention is not applicable.

As the Convention will need to be implemented at national levels, affected individuals will be protected by these domestic safeguards (given their state has signed on) and can seek legal remedies at domestic level. In addition, the Convention requires states to create a national supervisory authority to supervise the implementation of the Convention’s obligations. It does not include the possibility for individuals to claim violations of the new AI Convention directly before the European Court of Human Rights, whose mandate is limited to the European Convention on Human Rights (ECHR). However, individuals could still lodge complaints about a violation of their ECHR rights in relation to AI systems (on condition that national courts have rejected their complaint). The Strasbourg Court will then likely consider the principles enshrined in the specific legal instrument on AI.

What has happened so far?

• 2019 - 2021: The Ad hoc Committee on Artificial Intelligence (CAHAI) prepares the negotiations. It publishes a "Feasibility Study" on a legal framework on AI and adopts, in December 2021, its final recommendations on "Possible Elements of a Legal Framework on Artificial Intelligence Based on the Council of Europe’s Standards on Human Rights, Democracy and the Rule of Law."
• April 2022: The new Committee on Artificial Intelligence (CAI) holds its inaugural meeting in Rome and elects Ambassador Thomas Schneider (Switzerland) as its Chair. Its work will be informed by CAHAI’s final recommendations, but CAI is specifically tasked with the mandate to negotiate an "appropriate legal instrument."
• January 2023: Third plenary meeting of CAI. The Committee decides to split its meetings into meetings of the plenary and of the drafting group. The latter is charged with drafting proposals for the Convention’s text. Civil society organizations and other observers are excluded from participating in it.
• March 2024: In its final negotiation round, CAI agrees on a Draft Convention.
• April 2024: The Parliamentary Assembly of the Council of Europe (PACE) discussed the Draft Convention and adopts its "Opinion."

What are the next steps?

The AI Convention is due to be adopted by the Committee of Ministers on May 17, 2024. It will then be open for countries to sign. Signing is initially a declaration of intent - before the convention actually applies in a country, it must be "ratified" by that country. This requires the approval of national parliaments and the implementation of key provisions of the Convention at national level.

How is AlgorithmWatch contributing?

AlgorithmWatch participated as an active and official observer organization at CAI negotiations. Before that, we had contributed to CAHAI as an official observer in 2020 and 2021. The aim of our contribution had always been to ensure that the voice of civil society is heard in the negotiations in CAI – and to fight for a legal instrument on AI systems that is truly oriented at the Council of Europe’s mandate: the protection of our most basic rights, of our democracies, and the rule of law.

Now that the Convention has been adopted, we will work at national level in Switzerland and Germany to ensure that they implement the provisions courageously, comprehensively and with foresight - and not with a minimum solution.

• April 2024: We inform PACE members of our assessment of the Draft Convention, which they include in their statements.
• January & March 2024: Together with over 100 civil society organizations and researchers, we called on the negotiating States to cover the public and private sectors equally and to reject blanket exceptions for national security.
• January 2023: On the occasion of the International Day on Data Protection, we call for an effective Convention on AI.
• October 2022: Jointly with our partners, we urge the EU not to delay the process on a Convention on AI within the Council of Europe, given their different mandate.
• April 2022: At the opening session of CAI, AlgorithmWatch and its civil society partners reiterate their core demands.
• December 2021: At the end of CAHAI’s mandate and after two years of intense work, we are afraid that the CAHAI’s recommendations fall short of what is needed in terms of ensuring full respect for and protection of human rights, democracy, and the rule of law.

Read more on our policy & advocacy work on the Council of Europe.