AI Governance Engineer
GitHub

Harm at five levels

Five harm levels from one person to the environment, each with an example harm from the atlas and the control, by stack layer, that catches it.

By Updated First published

Harm at five levels Five harm levels from one person to the environment, each with an example harm from the atlas and the control, by stack layer, that catches it. Where harm lands, level by level Individual one person Discrimination in consequential decisions caught by Eval Gate in CI Group a group or community as such Exclusion through under-representation in data caught by Model Card as Control Evidence Organisation builds, buys or deploys Security compromise through prompt injection caught by Runtime Guardrail Society shared systems Job displacement and task transformation caught by Workforce impact assessment Environment the physical environment Energy demand and emissions of training and inference caught by Energy and emissions telemetry Harms atlas incidents, failure mode, control, evidence Chip colour: the control's layer Layer 01 Govern-as-Code Layer 02 Inventory & Transparency Layer 03 Evals & Red Teaming as Evidence Layer 04 Runtime Controls & Observability Layer 05 Assurance & Continuous Compliance
Harm at five levels Where AI harm lands, from one person to the physical environment, with one example harm per level from the harms atlas and the control that catches it, coloured by its stack layer. Calibrate your severity scale against every level, not only against harm to one person. Drawn from chapter 13.

Text alternative

Five levels at which harm lands, with one example per level from the harms atlas. Individual: harm to one person, their rights, money, liberty, autonomy or body; for example discrimination in consequential decisions, caught by the Eval Gate in CI (Layer 03 Evals & Red Teaming as Evidence). Group: harm that falls on a group or community as such, not only on its members one by one; for example exclusion through under-representation in data, caught by Model Card as Control Evidence (Layer 02 Inventory & Transparency). Organisation: risk to the organisation that builds, buys or deploys the system; for example security compromise through prompt injection, caught by the Runtime Guardrail (Layer 04 Runtime Controls & Observability). Society: harm to shared systems such as information, elections, work, equality and essential services; for example job displacement and task transformation, caught by a workforce impact assessment (Layer 01 Govern-as-Code). Environment: harm to the physical environment from building and running AI systems; for example energy demand and emissions of training and inference, caught by energy and emissions telemetry (Layer 05 Assurance & Continuous Compliance). The harms atlas holds every harm with real incident records, the failure mode an engineer can test for, the control that catches it and the evidence that control leaves.

Download

Every file carries the attribution band "aigovernanceengineer.com · CC BY 4.0 · v0.5.0" inside the image, and the version is in the file name, so a copy always says where it came from and which edition it shows. The SVGs keep the text live: the first follows the viewer's light or dark setting, the other two fix one theme for slides and print. The PNGs are drawn with the site's own typefaces.

Five harm levels from one person to the environment, each with an example harm from the atlas and the control, by stack layer, that catches it.
The PNG, light, 1600 px wide, as it downloads (shown here as a lighter copy).

Reuse and credit

The figure is published under CC BY 4.0: you may copy, share and adapt it, commercially too, provided you give appropriate credit, link to the licence and say if you changed it. Keep the attribution band in the image. A credit line that covers title, author, source and licence:

“Harm at five levels” by Jorge García Aibar, aigovernanceengineer.com (https://aigovernanceengineer.com/figures/harm-levels), v0.5.0. Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

Embed with HTML

<figure>
  <img src="https://aigovernanceengineer.com/downloads/figures/harm-levels-v0.5.0-light-1600.png" alt="Five harm levels from one person to the environment, each with an example harm from the atlas and the control, by stack layer, that catches it." width="800" height="1378" loading="lazy">
  <figcaption>
    <a href="https://aigovernanceengineer.com/figures/harm-levels">Harm at five levels</a> by Jorge García Aibar,
    aigovernanceengineer.com, v0.5.0.
    Licensed under <a href="https://creativecommons.org/licenses/by/4.0/">CC BY 4.0</a>.
  </figcaption>
</figure>

Embed with Markdown

![Five harm levels from one person to the environment, each with an example harm from the atlas and the control, by stack layer, that catches it.](https://aigovernanceengineer.com/downloads/figures/harm-levels-v0.5.0-light-1600.png)

*[Harm at five levels](https://aigovernanceengineer.com/figures/harm-levels) by Jorge García Aibar, aigovernanceengineer.com, v0.5.0. Licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).*