RAID & exposure
Risks that have an owner and a date
Exposure is probability × impact on a 1–5 scale. Anything scoring 20 or above belongs in the next steering conversation with a named response, not a colour.
3
Severe (20+)
2
High (12–19)
1
Moderate (6–11)
0
Low (1–5)
Exposure heat map
5
5
10
1
2
25
4
4
8
1
16
1
3
3
6
1
12
15
2
2
4
6
8
10
1
1
2
3
4
5
1
2
3
4
5
↑ impact
probability →
Log a new entry
Register · 6 entries
| Exp. | Type | Entry | Owner | Due | Status | |
|---|---|---|---|---|---|---|
| 20 | Risk |
Provider deprecates the model version in production
Atlas — Claims Triage · P4 × I5
Abstract the provider behind a routing layer; hold a tested fallback model and rehearse the swap quarterly.
|
M. Okafor | 2026-11-14 | ||
| 20 | Issue |
Cost per successful task is 7x the approved budget
Forge — Engineering Agent · P5 × I4
Introduce semantic caching and small-model routing for low-complexity requests; re-forecast before the scale gate.
|
K. Bhatt | 2026-10-24 | ||
| 20 | Risk |
Prompt injection via user-supplied documents leaks internal context
Lumen — Knowledge Assistant · P4 × I5
Untrusted-content isolation, output filtering and an adversarial red-team before any external exposure.
|
S. Alvarez | 2026-10-17 | ||
| 15 | Dependency |
Legal sign-off on training data provenance
Harbor — Contract Review · P3 × I5
Weekly standing slot with legal counsel; provenance register maintained by the data lead as a live artefact.
|
E. Petrova | 2026-10-31 | ||
| 12 | Risk |
Adoption stalls because support cannot explain AI behaviour
Atlas — Claims Triage · P3 × I4
Ship a failure-mode cheat sheet and run support enablement one wave ahead of each rollout cohort.
|
Priya Raman | 2026-11-21 | ||
| 9 | Assumption |
Frontline users will accept advisory-only output without override friction
Lumen — Knowledge Assistant · P3 × I3
Validate in the pilot via override-rate telemetry; if overrides exceed 30%, redesign the oversight model.
|
S. Alvarez | 2026-11-07 |