Respond
What is decided, by whom, and how robust the control is. 7 components, chapter 17 of the book.
Seven responses, named decider
Every risk carries exactly one response from accept, avoid, mitigate, share, transfer, realize, share, enhance; a row with no response is invalid. Alternatives are compared on cost, effectiveness and feasibility, and the register records the named person who decided and the date.
Hierarchy of controls
Every control carries its level (elimination, substitution, engineering, administrative, personal protection), a human-dependence flag and an expected decay profile; controls at the two lower levels are re-tested more often and their ratings age; aggregation weights by level, so a policy and an automated block never count the same.
Source: NIOSH / CDC [10]
Prevention and detection
Every control is tagged prevention (acts on the cause) or detection (acts on the mode); occurrence and detection ratings are revisited after any action with the previous values kept, so the register shows whether the action worked.
Source: ASQ FMEA [8]
Design gate
No new product, process, system or material change goes live before a review asks whether its hazards can be eliminated or substituted rather than administered; the answer, the level chosen and the signatory are recorded on the control.
Decision record
Responses above a stated materiality are recorded in a selection report: alternatives, tolerance per objective, risk of each alternative, rationale, expected exposure, decider, date. Acceptance of a risk above the appetite position is communicated to the governing body.
Proportionality
In the tolerable region, treatment continues until the next step is grossly disproportionate to its benefit; the disproportion is argued in writing with the cost and benefit of the next step, never assumed.
Source: risk-engineering.org [26]
AI controls
Every AI system is mapped, measured and managed against the seven trustworthiness characteristics with a current and a target profile; each carries decision rights set by appetite: act reversibly under monitoring, prepare for human approval, or inform only. Tested at model, adversarial and field level.