Sovereign Agents | CNAS

The report recommends that frontier AI developers name, evaluate and safeguard against sovereign agents in their safety frameworks. It also recommends that transparency and incident reporting legislation should require developers to do so, linking an internal developer practice to a statutory obligation.

The report makes a second recommendation addressed to developers and government bodies jointly. According to the excerpt, the purpose of those testbeds is to study how sovereign agents operate and how to control them. The excerpt describes the testbeds as 'maximally secure' and 'airgapped', and does not state where they would be built or who would fund them.

The excerpt does not define what a sovereign agent is, and it attaches no figures, timelines or cost estimates to either recommendation. What it sets out is a division of labour: developers would identify, evaluate and safeguard against sovereign agents within their own safety frameworks, and would be required to do so under transparency and incident reporting legislation; the airgapped testbeds would sit with the named developers, standards body and agencies for research into operation and control. No single regulator is named as responsible for oversight.