Are shopping and admin agents high-risk AI?
Last updated: · Leutrim Miftaraj
The honest answer: it depends on the use — the Annex III areas (employment, essential services, law enforcement and others) decide, not the technology. But the smart strategy does not hinge on the classification: whoever has oversight mechanics is well positioned in every scenario.
Take grey areas seriously
An HR agent (employment.hire) touches employment; an agent with actions towards authorities (permit.apply, commercialregister.filing) touches essential services. The taxonomy makes such contact points visible — risk level 3 signals where caution is needed.
No-regret architecture
Mandates, escalation, kill switch and evidence are good practice below the high-risk threshold and fulfil obligations above it. The investment is right in both worlds — the classification dispute becomes a footnote.
Transparency towards those affected
Regardless of classification: anyone affected by agent actions benefits from verifiable authority — the verification receipt is also an instrument of fairness.
Key takeaway
The classification question is for lawyers. The architecture question was settled long ago: build for oversight.
Try it yourself
Issue your first agent mandate and check your first decision with its receipt — free, in a few minutes.