As summarized by Claude:
- The Council of Europe is drafting an international treaty on AI called the Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law. It involves 46 European countries plus observers like the US, Canada, and Israel.
- The treaty was meant to align with the EU's AI Act legislative timeline, but has been delayed and is now expected to be finalized before the end of the year.
- The US is pushing to limit the treaty's scope to only public sector use of AI, excluding regulation of private companies. The EU opposes this. The US proposed an "opt-in" approach where countries can voluntarily extend the treaty to private sector.
- Other issues being debated are carve-outs for national security, weakening certain rights to 'principles', and limiting access to remedies.
- There are concerns the final treaty may be watered down to secure broader participation, especially if the US provides funding for outreach like it does for the Budapest Convention on cybercrime.
- Overall, the ambition for a strong global AI treaty is facing challenges from attempts to narrow its scope and enforcement mechanisms. The final outcome remains uncertain.
I wasn't able to easily find this story corroborated on other websites, but Euractiv is apparently a high-credibility website according to Media Bias/Fact Check.
Just as the timing of the world’s first AI treaty starts aligning with the EU legislative agenda, an American-led push to exclude private companies might make it not worth the paper it is written on.
The Council of Europe is an international body in charge of upholding human rights in its 46 participating countries, plus observers including the United States, Canada, Israel and Japan.
Since the start of 2022, all of these countries have been involved in the Committee on Artificial Intelligence, tasked with drafting the Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law.
For the EU, this binding international treaty could be a benchmark for its regulation of Artificial Intelligence, the AI Act.
However, Europe’s ambition to be the world’s digital rule-maker clashed with a practical problem: the work in the AI Convention had already started, whilst the EU regulation was still a moving target.
Last October, EURACTIV revealed that the European Commission, which would be entitled to negotiate on behalf of the EU, managed to buy some time. The timeline has been further delayed, and it is now fully aligned with the AI Act, set to be finalised before the end of the year.
The committee’s plenary session in September that was meant to endorse the final text of the AI treaty has been cancelled. The adoption at the committee level is now expected by March or April next year, followed by ministerial adoption in May or June.
However, precisely when the timing issue seems to be solved, another open front is getting in the way of the Commission’s ambition, notably that, upon pressure from Washington, the whole treaty might be watered down.
Crack-down on transparency
In January, EURACTIV revealed that the US had requested and obtained that the drafting of the text would occur behind closed-door meetings where only the potential signatories, meaning the country representatives, could participate.
The result is that stakeholders like civil society groups and private actors were left out in the cold until the following plenary. Only one plenary session is expected in late October before adoption, whereas four or five informal drafting meetings are planned to discuss the outstanding controversial points.
The reason for such secrecy, which runs against the CoE’s internal policy, comes down to the fact that the participating countries did not want their positions publicly known, especially since the US had been pushing to curb the scope of the treaty as of day one.
With the support of the United Kingdom, Canada and Israel, the US government has been pushing to limit the scope AI Convention to only public bodies, leaving out the private sector. By contrast, the committee’s mandate refers to a “binding legal instrument of a transversal character.”
The opt-in option
Despite opposition from the EU, the possibility of excluding private companies remains on the table. The US administration has refined its proposal, presenting an opt-in option during a plenary committee session last week in Strasbourg.
“Any Party may, at the time of signature, when depositing its instrument of ratification or accession or any time after that, declare the extent to which this Convention shall apply within its jurisdiction, beyond the design, development and use of artificial intelligence systems by or on behalf of the Party,” reads the US proposal, seen by EURACTIV.
In other words, the Convention would only apply to public bodies. Still, the individual country could decide whether the international treaty would also apply to the private companies in its national jurisdiction and publicly declare it.
This approach is somewhat unusual. Normally, if the treaty allows it, the countries would declare if they have reservations or opt out from certain parts. However, an opt-in version would spare the government significant public embarrassment.
Albeit the American proposal would mean lowering the AI Convention below the Council of Europe standards, EURACTIV understands that the committee’s secretariat has so far indulged the idea for the sake of securing as broad participation in the treaty as possible.
According to a source informed on the matter that spoke to EURACTIV under the condition of anonymity, one of the reasons for this lenient approach might be related to the fact that the Council of Europe is hoping that Washington will finance outreach initiatives related to the AI treaty, much like it is currently doing for the Budapest Convention on cybercrime.
Other open fronts
Other fronts might result in weakening the first international treaty on AI. While national defence has already been excluded from the current draft, participating countries are also discussing a broader carve-out for national security.
Another critical issue is that certain governments are pushing to change rights recognised by the Council of Europe, like data protection and non-discrimination, to more general ‘principles’, which would leave the countries much more room for interpretation.
Finally, also controversial is how robust needs to be the follow-up mechanism and whether it should involve external experts. A fundamental part of this chapter is that some countries would like discretion in restricting access to remedies.