The Council of Europe’s Framework Convention on AI was agreed on Thursday in Vilnius, Lithuania, marking a major step in aligning AI development with human rights, democracy, and the rule of law.
The United States, European Union, United Kingdom, and Israel were among the initial signatories.
Other signatories included Andorra, Georgia, Iceland, Norway, the Republic of Moldova, and San Marino. More countries are expected to join over the coming weeks and months.
Developed over two years with input from more than 50 countries, it aims to create a unified global, legally binding framework for AI governance.
A new chapter in AI regulation
This treaty is the latest in a series of efforts to regulate AI.
Recent months have seen a flurry of agreements, including the EU’s AI Act, the G7 deal last October, and the Bletchley Declaration signed by 28 countries in November 2023.
In May, 16 international tech companies, including Amazon, Google, and Microsoft, also signed a new voluntary safety framework.
AI governance initiatives are stacking up high. It reflects mounting concern surrounding AI’s potential risks.
However, to date, many have doubted whether voluntary commitments have any ‘teeth.’ Do they amount to more than sentiment?
The Council of Europe’s convention aims to address this with a legally binding treaty. It requires signatories to implement safeguards against AI-related threats to human rights, democracy, and the rule of law.
The treaty mandates accountability for harmful outcomes and ensures AI respects privacy rights.
Věra Jourová, vice president of the European Commission for Values and Transparency, called the convention a blueprint for global AI regulation: “This framework should build trust that AI innovations respect our values.”
The treaty covers AI use in the public and private sectors. Organizations must assess their AI systems’ impact on rights and democracy and make their findings public.
Further, users must know when they’re interacting with AI, not humans, and can challenge AI-made decisions.
While described as “legally enforceable,” the framework seems to lack specific sanctions like fines. It relies primarily on monitoring for compliance.
There have been bits and pieces of enforcement against AI companies, including this week when Clearview AI was fined $33m by a Dutch agency for scraping facial images for its image database.
Nevertheless, it’s another step towards unifying global AI rules and complements the AI Act by introducing enforceable rules outside of the EU.
The convention’s global scope is key. Council of Europe Secretary General Marija Pejčinović Burić expressed hope for “many signatures and quick ratifications.”
The treaty needs to be ratified by five signatories, including three Council of Europe members, for it to take effect.
Once in force, it will open to countries worldwide, potentially creating a truly global standard for AI governance.
