§5 · Lane 2 — AI Audit and Accountability
Closing the AI Accountability Gap (SMACTR)
Raji et al. (2020) · FAT* '20
Bibliographic data
- Title
- Closing the AI Accountability Gap: Defining an End-to-End Framework for Internal Algorithmic Auditing (Raji et al., 2020)
- Authors / Issuing body
- Inioluwa Deborah Raji, Andrew Smart, Rebecca N. White, Margaret Mitchell, Timnit Gebru, Ben Hutchinson, Jamila Smith-Loud, Daniel Theron, Parker Barnes
- Venue / Publisher
- Proceedings of the 2020 ACM Conference on Fairness, Accountability, and Transparency (FAT\* '20), Barcelona, January 2020, pp 33-44
- Year
- 2020
- Designation
- Academic
- Licence
- DOI — refer to publisher for full licence terms.
- Canonical link
- https://doi.org/10.1145/3351095.3372873
How to cite
Raji et al. (2020). Closing the AI Accountability Gap: Defining an End-to-End Framework for Internal Algorithmic Auditing (Raji et al., 2020). Proceedings of the 2020 ACM Conference on Fairness, Accountability, and Transparency (FAT\* '20), Barcelona, January 2020, pp 33-44. https://doi.org/10.1145/3351095.3372873.
Introduces the SMACTR framework (Scoping → Mapping → Artifact Collection → Testing → Reflection) for internal algorithmic auditing — an end-to-end audit process applied throughout the AI system lifecycle rather than after deployment.
Why it matters for NETEVO
Raji and colleagues' Closing the AI Accountability Gap is the founding citation of the AI-internal-audit literature, and one of the load-bearing academic anchors for NETEVO's engineered-governance thesis. The paper introduces SMACTR — Scoping, Mapping, Artifact Collection, Testing, Reflection — an end-to-end internal audit process executed across the AI system lifecycle rather than retrofitted after deployment.
Audit integrity depends on procedural justice. The paper's central argument is that an audit's authority derives from the procedure under which it was conducted, not from the seniority or independence of who conducted it. This is the academic foundation for NETEVO's position that paper governance fails an audit, and engineered controls survive one — engineered controls survive because a vetted procedure can be executed against them objectively, whereas paper policies and board minutes cannot be tested in the same way.
Internal audit during development, not after deployment. SMACTR maps audit activity through every stage of the AI system lifecycle, treating audit as a build-time discipline rather than a post-hoc review. NETEVO's Law-to-Code Methodology is the same shape: obligations identified by counsel or compliance are encoded into executable controls during build, not reconstructed from evidence after the fact. The paper also imports audit-integrity practices from aerospace, medicine, and finance into AI — exactly the cross-domain rigour NETEVO claims as a differentiator over generic AI-governance advisory work.
This is why SMACTR is the load-bearing primary citation for the forthcoming AI-Washing Audit whitepaper, and why NETEVO cites it on the About / Methodology page as the procedural-justice foundation of the engineered-governance argument.