Systems that, given a goal, plan and execute actions in other tools: sending emails, purchasing, changing records or approving requests. Their risk depends on the permissions they hold and whether they decide about people without human involvement.
We only list countries where an obligation in force or adopted mentions this use case.
| Jurisdiction | Level | Date | Note |
|---|---|---|---|
| European Union | Transparency | August 2, 2026 | Interaction notice if it deals with people; if the agent decides in an Annex III area, it will be high-risk from Dec 2, 2027; the draft guidelines treat multi-agent systems as one system. |
| Spain | Data protection | — | The AEPD issued guidance on agentic AI (2026): memory, autonomy, permissions and GDPR Art. 22 if the agent decides about people. |
| Peru | Minimal risk or no specific regime | September 10, 2026 | Acceptable risk unless it acts in a high-risk area. Deadline depends on the company's sector: health, education, justice, security and finance Sep 10, 2026; transport, commerce and labor Sep 10, 2027; production, agriculture, energy and mining Sep 10, 2028; others Sep 10, 2029; small businesses Sep 10, 2027 and micro businesses Sep 10, 2028 unless their sector has an earlier date. |
| United States (federal) | Minimal risk or no specific regime | — | No specific law; banking guidance SR 26-2 excludes agentic AI from model risk, leaving it to general governance. |
Policy · USD 39 · Available to buy
Procedure · USD 39 · Available to buy
Protocol · USD 39 · Available to buy
Register · USD 59 · Available to buy
Procedure · USD 39 · Available to buy
Includes 8 documents
Six base documents for agents that act in systems: autonomy and permissions policy, integration map, adversarial testing, human oversight, rollback and incident response.
USD 169USD 292
Complete kit · available to buy