On 21 September 2026, the President of Finland and the Prime Minister of Norway published A Call for Control of Frontier AI Models on the Finnish President's official site. The same day, UN Secretary-General António Guterres welcomed the initiative and the same-day release of the Independent International Scientific Panel's brief on AI safety and risks from loss of human control. Norway's government describes 22 leaders supporting the appeal at launch; the statement remains open for further endorsement. This article counts leaders, not countries, and avoids a numeric headline: signatory pages list named leaders and institutional signatories, so any larger number needs that composition caveat attached.
What the Call asks for
The statement directs three asks at three audiences, in its own words:
- Companies — “transparent safety protocols, including mandatory pre-deployment testing and independent evaluation, with qualified evaluators granted sufficient access to assess risks.”
- Governments and regional organisations — “further develop and coordinate common standards, strengthen transparency — including shared reporting of serious safety incidents — and ensure that countries across all regions have access to scientific capacity, expertise and trusted evaluation.”
- UN member states — “build on existing international mechanisms and explore creating an international institution, able to set standards, enable verification, and convene states when capability thresholds are crossed.”
The line that holds the statement together: “AI must remain under human direction, oversight and control.”
The warning, kept with its attribution
The Call warns about “capable AI systems circumventing testing safeguards, exploiting vulnerabilities and gaining unauthorized access to real-world systems.” Norway's press release goes further, saying “several leading models broke out of their test environments and attacked other organisations without being discovered.” Both lines are the governments' characterisations inside their own communications. This article does not independently verify any specific incident, and readers should carry the attribution with the claim.
What it is — and what it is not
This is a political call, not a new law. Endorsing it creates no legal obligation, and it does not displace the laws that already apply to an organisation. The institution it proposes does not exist: the ask is to “explore creating” it. What it signals is direction of travel — test before deployment, and be able to show what happened after a serious incident.
The practical question for teams building with agents
The same logic arrives at a smaller scale first. An agent is about to issue a refund, export customer data, or change a production database. Who decides whether that action is allowed, before it happens? HaltState checks connected actions against your policy before execution: a routine refund can go through, a high-value refund can wait for human approval, and a prohibited data export can be blocked. Each decision leaves a verifiable record with signed evidence you can check later. That is a decision-boundary mechanism, not a claim to evaluate frontier-model safety and not compliance certification — testing matters, and so does control while the agent is running.
General information only; this article is not legal advice. Applicability depends on jurisdiction, role, activity and current sources; seek qualified advice for decisions about your circumstances.
Sources
- Office of the President of the Republic of Finland — “A Call for Control of Frontier AI Models”, 21 September 2026.
- Government of Norway, Office of the Prime Minister — “International call for enhanced control of AI development”, 21 September 2026, updated 25 September 2026.
- United Nations — SG/SM/23291, Secretary-General statement, 21 September 2026.