ROW EDITION · updated Aug 8, 2026
HARMS worksheet
Name the humans, plausible harm, defense, owner, and receipt before automation gets to act.
← All projector / phone kit pages
Open this page on your phone
HARMS an always-on agent
Scenario: an agent reads mail, watches chat, schedules meetings, and acts without being asked — including during an emergency.
H — Humans affectedWhose real lives could be changed? Name people, not categories.
A — Adverse outcomeWhat is the worst plausible outcome, not the worst imaginable?
R — Risk defenses in placeWhat stands between today and that outcome? What guardrail was lowered?
M — Mitigation & ownerWhat defense will you add, and which named role owns it?
S — Signal & receiptWhat will prove the defense works? What record will you keep?
Highest-risk decision: __________________________________________
Mitigation we will actually run: __________________________________________
Mitigation we will actually run: __________________________________________