“We need to create a sustainable legal framework for AI that reaches across our continent—and even beyond. A new framework convention that can be a platform for co-operation and set important standards, just like the Budapest Convention. And I hope that this is what we can now work towards.”
These optimistic words came from Bjørn Berge, Deputy Secretary General of the Council of Europe at the Inaugural meeting of the Committee on Artificial Intelligence back in April 2022. At the time, two major European efforts to address AI governance—the European Union’s Artificial Intelligence Act and the Council of Europe’s process to elaborate a legal framework on artificial intelligence—were both running smoothly in parallel.
Many hope (and expect) that these parallel efforts—particularly at the Council of Europe—will have influence beyond Europe, and help drive uniform standards and safeguards around human rights and AI. But, in September, an obstacle emerged. During CAI’s second plenary meeting, the EU Commission unexpectedly issued a statement, expressing the need for more time to ensure coordination and alignment between the proposed zero draft CoE Convention on AI and the EU’s own AI Act. This effectively barred EU member states from negotiating as individual states until the EU Commission has received a mandate from the EU Council to negotiate on their behalf. As a result, the next CAI plenary meeting—originally set to take place this month—will now be held in 2023 while the EU Council finalises the mandate for the Commission and the other areas in which Member States will be allowed to negotiate on their own.
This move was greeted with widespread disappointment from human rights defenders—including civil society observers at CAI.
Below, we set out three reasons why this stance and a sustained delay to the negotiating process would be a missed opportunity.
1. It might mean non-EU states have to wait longer for a binding instrument on AI
Delays to negotiations may negatively impact non-EU member states that are not subject to the EU AI Act when it is passed, even if just temporarily. This is because the EU AI Act will only apply to EU member states and to non-EU member states when an AI system has an impact within the EU internal market.
Unlike the EU AI Act, all countries will be invited to join the CoE’s Convention on AI. It presents an opportunity to set a binding global standard on AI governance which protects human rights, democracy and the rule of law. There is precedent for this as states outside the Council of Europe have become parties to other CoE instruments; namely, the Budapest Convention. For several decades now, the Budapest Convention has remained the only international treaty on cybercrime, and its membership has grown to include countries outside the Council of Europe, including the US, Canada, Japan, Israel and others. These countries are important global actors, are currently active in the CAI process, and would be potential parties to the convention. Delays to the negotiations could theoretically set back the date the convention ultimately becomes effective, and could potentially risk its legitimacy or normative value.
2. It may hinder progress made by CAI (and its predecessor CAHAI) to protect human rights, democracy and the rule of law
Will the work of CAI and final convention adequately protect human rights? At present, it isn’t possible to definitively say that: it is still being elaborated and will be subject to further negotiations, so there is a lot still to be decided.
But there are some crucial indications that it might. The government ministries participating in CAI are those responsible for the protection of fundamental rights in their countries. Moreover, the final output of CAHAI noted that a legally binding instrument “should focus on preventing and/or mitigating risks emanating from applications of AI systems with the potential to interfere with the enjoyment of human rights” and should be underpinned by a risk-based approach, with with basic principles that are applicable to all AI systems.
And a recently released document from the CAI—entitled Outline of the Elements of an Appropriate Legal Instrument Proposal from the Secretariat—further supports this approach. It suggests that the proposed legal instrument would set out principles to address known issues which might arise with the use of AI “such as those pertaining to potential bias and unlawful discrimination, data protection and data governances as well as issues of human agency, accountability and responsibility”. The document also references the use of AI by the public sector “where special principles, rules and possibly obligations would be necessary to address any concerns regarding respect for human rights, observance of democracy and the rule of law.” These texts indicate that significant obligations, protections and safeguards could be included within the final convention.
These are critical given ongoing criticisms of the proposed EU AI Act as being insufficient to provide sufficient protections for human rights. As tweeted by Director Jan Kleijessen of the Council of Europe, “It is exactly because of the EU’s #AIACT focus on market issues and only parietal consideration of #HumanRights that @COE4AI negotiations on a possible global Convention on #AI, #HumanRights, #RuleOfLaw and #Democracy are so essential. In parallel, complementary and urgent.”
3. We can’t afford to wait
The next CAI plenary meeting has already been delayed until January 2023 as a result of the EU’s announcement. It is unclear whether more delays are coming and how CAI will be able to move forward with negotiations given ongoing developments, including proposals by the Commission to discuss less controversial elements of the instrument.
What is clear is that delays to these negotiations could result in a prolonged absence of protections and safeguards. This issue has been raised by civil society organisations that participate as observers at CAI. In the recent joint statement, they noted the opportunity for non-CoE member states to adopt the Convention in the future means that “…the impact of the Convention to protect fundamental human rights, rule of law, and democratic institutions can be magnified. A delayed process and/or a weakened Convention would be a setback globally”. Similar concerns have also been shared by those active in CAI and its predecessor CAHAI, including Gregor Strojin, vice-chair of the Committee on AI, who has gone on the record to say that “No regulation is regulation already. Until binding and effective legal instruments are adopted and applied, unacceptable patterns and practices will entrench themselves further.”