Cross-reference
NIST AI 600-1
12 items: 9 met directly by a control at Level 3, 3 partly, 0 deliberately out of scope.
| Ref | Item (our description) | Controls | Note |
|---|---|---|---|
| R1 | CBRN Information or Capabilities; AI making it easier to get or design chemical, biological, radiological or nuclear weapons knowledge | RSK-02 partial, RSK-09 partial, GOV-01 partial, MON-02 partial, SEC-02 partial | Held at partial: RSK-02 requires the method to cover the recognised generative AI risk categories, which names CBRN only generically; no control addresses this risk specifically. Supporting controls are listed. Revisit if a specific control is added. |
| R2 | Confabulation; confident but false or invented output, including made up citations and reasoning | TEC-06 direct, MON-02 partial, RSK-07 partial, PPL-01 partial, OPM-09 partial, MON-04 partial | Accuracy testing before release (TEC-06) and accuracy monitoring in live use (MON-02) identify it; human review (RSK-07), literacy about limits (PPL-01), fallbacks for wrong output (OPM-09) and incident reporting (MON-04) manage it. Checking facts and citations in general purpose tools used by staff rests on literacy and oversight rather than testing. |
| R3 | Dangerous, Violent, or Hateful Content; AI producing inciting, violent, self harm or hateful output, including when safeguards are bypassed | TEC-06 direct, SEC-06 partial, RSK-09 partial, RSK-02 partial, MON-02 partial, SEC-02 partial, GOV-01 partial, RSK-03 partial, MON-04 partial | TEC-06 Level 3 requires generative AI testing to cover harmful output, SEC-06 tests attempts to bypass safeguards, and RSK-09 assesses supplier content safeguards. Monitoring (MON-02), threat modelling (SEC-02), prohibited uses (GOV-01), safety impact (RSK-03) and incident handling (MON-04) contribute. Filtering harmful output in live use is not named. |
| R4 | Data Privacy; leakage, misuse or inference of personal and sensitive data through AI training, prompts and outputs | DAT-05 direct, SEC-04 direct, RSK-03 direct, DAT-07 partial, DAT-08 partial, RSK-09 partial | Lawful and fair use (DAT-05) and leakage rules with technical measures (SEC-04) carry it, and RSK-03 privacy impact names sensitive inferences about people. TEC-06 tests that personal or sensitive information is not disclosed and MON-02 monitors outputs for it in live use. |
| R5 | Environmental Impacts; energy, carbon and other resource costs of training and running generative models | RSK-02 partial, OPM-02 partial, RSK-09 partial, RSK-03 partial, MON-01 partial | Held at partial: RSK-02 requires the method to cover the recognised generative AI risk categories, which names environmental cost only generically. Supporting controls are listed. Specific wording: MON-01 measures environmental footprint (energy, compute, water) where material, with a baseline, OPM-02 considers reuse of existing and open models before training new ones, and RSK-09 covers supplier sustainability practice. |
| R6 | Harmful Bias or Homogenization; AI amplifying bias, performing worse for some groups or languages, or giving overly uniform output | RSK-06 direct, TEC-06 partial, MON-02 partial, DAT-03 partial, RSK-03 partial, TEC-07 partial | Bias across groups is carried by RSK-06, with fairness measures monitored in live use through MON-02. TEC-07 records concentration on a single foundation model or supplier as a dependency risk; overly uniform output is not named. |
| R7 | Human-AI Configuration; people trusting AI too much or too little, treating it as human, or becoming emotionally reliant on it | RSK-07 direct, MON-01 direct, RSK-03 partial, PPL-01 partial, PPL-05 partial, CUL-07 partial, MON-06 partial | Designed oversight with trained reviewers (RSK-07) addresses automation bias in decisions; literacy and role learning (PPL-01, PPL-05) and adoption support (CUL-07) address over and under reliance, and MON-01 measures reliance against accuracy for each use so over- and under-reliance show up. RSK-03 covers over-reliance and people treating the AI as human for public-facing assistants. |
| R8 | Information Integrity; AI easing the spread of misleading content, synthetic media and impersonation | GOV-10 direct, SEC-02 partial, RSK-08 partial, GOV-09 partial, TEC-02 partial, TEC-06 partial | GOV-10 Level 3 sets rules for reviewing and disclosing AI generated content released outside the organisation, including synthetic media; SEC-02 assesses impersonation of the organisation or its people through synthetic media; RSK-08 tells people when they interact with AI and TEC-02 marks generative outputs where the organisation offers generative AI. |
| R9 | Information Security; attacks on AI such as prompt injection and data poisoning, and AI lowering the barrier to cyber attack | SEC-02 direct, SEC-01 direct, SEC-03 partial, SEC-04 partial, SEC-05 partial, SEC-06 partial | Attacks on the organisation's own AI are carried by the security baseline (SEC-01) and AI threat modelling (SEC-02). Attackers using generative AI against the organisation (better phishing, faster exploitation) is a general cyber security matter outside this framework's scope. |
| R10 | Intellectual Property; AI reproducing protected work, exposing trade secrets, or copying likeness or voice without permission | OPM-04 direct, DAT-08 direct, RSK-04 partial, SEC-04 partial, GOV-01 partial | Contract terms on IP in inputs and outputs (OPM-04) and rights over training data (DAT-08) carry it; leakage rules (SEC-04) protect trade secrets. No control asks staff to check generated output for infringement before it is published or reused. |
| R11 | Obscene, Degrading, and/or Abusive Content; AI producing abusive imagery such as synthetic child abuse material or non consensual intimate images | RSK-02 partial, RSK-09 partial, GOV-01 partial, MON-02 partial, MON-04 partial, TEC-06 partial, DAT-08 partial | Held at partial: RSK-02 requires the method to cover the recognised generative AI risk categories, which names abusive content only generically; no control addresses this risk specifically. Supporting controls are listed. Revisit if a specific control is added. |
| R12 | Value Chain and Component Integration; untraceable or poorly vetted third party models, data and components in AI systems | RSK-09 direct, SEC-03 direct, DAT-08 direct, OPM-04 partial, TEC-07 partial, GOV-06 partial |
References and short descriptions only; see the source publication for its own text. The cross-reference is the author's interpretation and is not endorsed by the publisher. See notices.