A group of 22 countries has backed a declaration calling for AI to remain under human control, adding another layer to the growing international debate over AI governance.

The declaration was discussed during the 2026 UN General Assembly. The United States and China were not among the signatories.

What does the declaration call for?

The declaration emphasizes human intervention and stronger safety practices.

It calls for AI companies to develop transparent safety protocols, including:

Pre-deployment testing

Independent evaluations

Greater transparency

Human intervention

International cooperation

The initiative comes amid increasing attention to AI agents that can act autonomously.

Why are signatories focusing on human control?

AI systems are becoming capable of performing tasks rather than simply generating information.

That changes the governance challenge.

A chatbot generating an incorrect answer is one problem.

An autonomous agent making a financial transaction, modifying software or accessing sensitive infrastructure creates a different one.

Human control therefore needs to be reflected in system architecture.

What does that look like technically?

Organizations can implement human control through mechanisms such as:

Permission boundaries: Agents receive only the access they need.

Approval thresholds: Sensitive actions require human authorization.

Audit logs: Every important action is recorded.

Kill switches: Agents can be stopped quickly.

Monitoring: Unexpected behavior triggers alerts.

These mechanisms allow organizations to benefit from automation without giving AI unlimited authority.

What happens next?

The declaration itself does not create a global AI regulator.

It represents a political and policy commitment from the participating countries.

The United States and China's absence also shows that achieving broad international agreement will be challenging.

For businesses, however, the principle is already practical.

AI systems should be designed so that humans can understand, supervise and intervene when necessary.

As agents become more capable, human control will increasingly become an architectural requirement rather than simply a policy statement.