Risk Register
Excerpt: 8 of 31 risks in the full register.
InherentHigh 4Medium 4Low 0ResidualHigh 0Medium 5Low 3Needs COO sign-off 5In treatment 1Treated (monitoring) 7
Heat map
Likelihood
5
4
3
2
1
12345Overall impact
Inherent
| ID | Risk title | AI system | Inherent | Residual | COO sign-off | Status | Owner |
|---|---|---|---|---|---|---|---|
| R-01 | Proxy features lead to discriminatory attrition predictions | SYS-02 Kestrova Retain | 20 High | 8 Medium | Yes | Treated (monitoring) | VP Engineering |
| R-02 | Scores used for redundancy selection (misuse) | SYS-02 Kestrova Retain | 15 High | 10 Medium | Yes | Treated (monitoring) | Head of Product, AI |
| R-03 | Applicants rejected with no human review | SYS-01 Kestrova Shortlist | 20 High | 10 Medium | Yes | Treated (monitoring) | Head of Product, AI |
| R-04 | Transcription errors bias interview scores | SYS-03 Kestrova Interview Notes | 12 Medium | 6 Low | Yes | Treated (monitoring) | Head of Product, AI |
| R-05 | Silent LLM model upgrade changes behaviour | All three systems | 12 Medium | 3 Low | No | Treated (monitoring) | VP Engineering |
| R-06 | Customer HR data used for training without clear permission | SYS-02 Kestrova Retain | 15 High | 10 Medium | No | In treatment | General Counsel (contract); VP Engineering (register) |
| R-07 | LLM vendor retains or reuses prompt data contrary to assumptions | All three systems | 8 Medium | 4 Low | No | Treated (monitoring) | VP Engineering |
| R-08 | Recruiters over-rely on LLM candidate summaries that contain errors | SYS-01 Kestrova Shortlist | 12 Medium | 8 Medium | Yes | Treated (monitoring) | Head of Product, AI |