Netherlands signs international declaration on AI safety
Governments, AI companies and civil society organisations agree on joint model tests and publication of non-sensitive test results.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
The Netherlands has signed an international declaration on the safety and security of advanced AI, together with five other countries and several AI companies. According to the declaration, the agreements focus on joint testing, transparency and rapid warnings, but for now mainly constitute a voluntary framework for cooperation.
The declaration was signed on Monday by Bahrain, Canada, Finland, France, Norway and the Netherlands. Companies including Anthropic, Cohere, Hugging Face, Microsoft, Mistral AI, OpenAI and Synthesia also support the text, as do several civil society organisations.
The signatories say that advanced AI models are being developed and used across borders. They therefore want to establish common methods for testing models and assessing risks. These include misuse for cyberattacks, chemical, biological, radiological or nuclear applications, and risks arising as models become more autonomous.
One of the agreements is that non-sensitive results from safety tests should be made public in a standardised format. The declaration also refers to a publicly accessible joint repository containing, among other things, protocols for red-teaming, benchmark tests and stress-test results.
The signatories also discuss responsibility throughout a model’s entire life cycle. Developers should explain capabilities, limitations and known risks and, where appropriate, allow independent evaluation. They also mention safe shutdown mechanisms, continuous monitoring and protection for employees who report serious risks.
The declaration is not a treaty and contains no detailed enforcement mechanism. The text also remains open to new signatories. For Dutch companies, the framework could become relevant if the proposed standards are later linked to European rules, public procurement or oversight. Its practical significance will depend on which agreements are actually converted into technical standards and national or European obligations.
Also show where in Europe and in the world
One story, several perspectives
What is established
- Six countries, including the Netherlands, signed the declaration.
- Several AI companies and civil society organisations support the text.
- The agreements cover evaluation, transparency, incident reporting and risk management.
- The declaration contains no detailed binding enforcement regime.
Left
Arguments AI companies should not decide for themselves how safety is measured; independent assessment, public transparency and the protection of citizens should come first.
Values Public oversight, equality, fundamental rights and prevention of social harm.
Consequences Stricter standards could limit risks, but could place additional burdens on smaller developers and open research.
Centre
Arguments International standards are needed because AI operates across borders, but agreements must remain technically feasible and compatible with existing regulation.
Values Proportionality, institutional cooperation and innovation with safeguards.
Consequences A common testing framework could make oversight more efficient, provided countries later develop the agreements in legal and technical terms.
Right
Arguments Governments should improve safety without unnecessarily restricting innovation and competitiveness; companies need predictable rules rather than divergent national regimes.
Values Freedom to do business, technological progress and national economic strength.
Consequences Light-touch and predictable rules could encourage investment, but voluntary agreements may provide insufficient protection in the event of serious incidents.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The signatories and substantive commitments were checked against the official declaration. The text explicitly states that it is a declaration rather than binding legislation.
- confirmed The Netherlands signed the declaration on 9 November 2026. — The Canadian government website lists the Netherlands among the signatory countries. source
- confirmed The declaration refers to joint model tests, standardised publication of non-sensitive results and a public repository. — These agreements are set out in the text of the declaration. source
- confirmed The declaration is not a treaty with a detailed enforcement mechanism. — The text presents voluntary commitments and contains no treaty or sanctions provisions. source
Editor's note
The declaration contains political and voluntary commitments, not new binding legislation. Its practical implementation and enforcement have not yet been established.Sources
- Joint declaration on Paris call for AI safety and security — Prime Minister of Canada
- G7 Foreign Ministers Meeting in Paris — Federal Foreign Office of Germany
More on this in Dutch media
- NOS — „kunstmatige intelligentie”
- Het Parool — „kunstmatige intelligentie”
- NRC — „kunstmatige intelligentie”