AI Governance Institute
← News
Research2026-09-12

IEEE Survey Links Explainability and Fairness as a Single Audit Obligation

Source

On the Interplay of Explainability and Fairness in AI: A Survey

IEEE Computer Society

Via IEEE Computer Society

What happened

The IEEE Computer Society published On the Interplay of Explainability and Fairness in AI: A Survey, a peer-reviewed synthesis of research on how explainability and fairness interact in machine learning systems. The survey finds that neither property can be fully evaluated in isolation: detecting and correcting bias typically requires explanation of how a model reached its outputs, while producing reliable explanations often depends on a model that treats demographic and other sensitive attributes consistently. For compliance teams, this framing reframes bias documentation not as a standalone exercise but as one that must be integrated with explainability infrastructure. The survey is relevant across jurisdictions where regulators now require evidence of both fairness testing and model transparency, including under the EU AI Act: High-Risk AI Systems, Transparency, and Enforcement Powers Applicable 2 August 2026 and frameworks such as the NIST Artificial Intelligence Risk Management Framework Playbook.

Why it matters

  • ·Regulators under the EU AI Act: High-Risk AI Systems, Transparency, and Enforcement Powers Applicable 2 August 2026 expect documented evidence of both transparency and non-discrimination for high-risk AI systems. Treating these as separate audit tracks risks leaving gaps that enforcement actions can exploit.
  • ·Model validation teams that scope bias testing and explainability reviews independently may be running incomplete assessments. The survey's central finding -- that bias can only be diagnosed reliably when the model's reasoning is legible -- means a weak explainability program is also a fairness control failure.
  • ·Organizations facing hiring, credit, and benefits decisions are the most immediately exposed, given that automated decision tools in these contexts must satisfy both bias and transparency obligations under rules such as the Proposed CPPA Regulations on Cybersecurity, Risk Assessments, and Automated Decision-Making Technologies and comparable state-level frameworks. Auditors presenting documentation that treats these as parallel but disconnected tracks may find their evidence packages rejected.

Governance controls affected

ALC-004AI Explainability DocumentationMON-003AI Bias and Fairness MonitoringMGV-006RAI Benchmark-Aligned Evaluation FrameworkHOC-001AI System Risk ClassificationALC-002High-Risk AI Audit Trail

What to do now

  • ☐Review your model validation methodology to confirm that bias testing and explainability review are conducted as integrated, not parallel, workstreams.
  • ☐Update audit documentation templates to require explainability evidence as a prerequisite for signing off on fairness assessments in high-risk AI systems.
  • ☐Map your current explainability outputs against fairness metrics for any model subject to the EU AI Act high-risk provisions or U.S. automated decision-making regulations, and identify documentation gaps.
  • ☐Brief model risk and validation leads on the survey's key finding: that unexplainable models cannot produce reliable bias assessments, and factor this into model approval gate criteria.
  • ☐Include the explainability-fairness interdependency as a standing agenda item in your AI governance committee's model review cadence, particularly for models used in lending, hiring, benefits, or healthcare decisions.

What to watch next

Regulatory guidance under the EU AI Act: High-Risk AI Systems, Transparency, and Enforcement Powers Applicable 2 August 2026 will increasingly treat unexplainable outputs as a proxy for inadequate bias controls, meaning enforcement scrutiny is likely to converge on both dimensions simultaneously. The NIST AI 600-1 Generative AI Profile and related NIST outputs are also expected to address explainability and fairness in generative settings, where the interdependency the survey describes is harder to operationalize. Compliance teams should monitor whether auditing standards bodies and sector regulators -- particularly in financial services and healthcare -- begin requiring integrated explainability-fairness evidence packages as a baseline submission standard.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Powered by Buttondown.

Related Coverage

Research2026-09-01

SR 26-2 Forces Banks to Rethink Model Governance From Inventory to Board Oversight

The OCC and Federal Reserve's revised model risk management guidance, SR 26-2, resets supervisory expectations for U.S. banks by shifting to a materiality-based approach that covers both traditional statistical models and AI systems, replacing the SR 11-7 framework that had governed bank model governance since 2011. Practitioner analysis from CRA identifies four areas banks must redesign: inventory scope, model tiering, validation independence, and governance alignment up to the board. A companion implementation guide from Lumenova AI adds concrete steps, including inventory rationalization and a distinct governance lane for agentic and generative AI, while a proposed academic framework maps a six-layer control architecture for bringing GenAI systems into SR 26-2 scope. Banks that still run AI governance and model risk management as separate programs face the most immediate pressure to harmonize them.

Research2026-09-11

25 Fields Medalists Warn AI Math Benchmarks Erode Attribution and Auditability

A declaration signed by 25 Fields Medal recipients, published by mathandai.org, argues that AI companies using mathematical problem-solving as a benchmark are fundamentally misaligned with the goals of the mathematical community. The signatories contend that AI-generated proofs are rapid and unreferenced, threatening attribution norms and disrupting the knowledge transmission chains that make mathematical research meaningful. The declaration frames these concerns as a governance signal relevant to any knowledge-intensive domain where AI outputs are replacing traceable human reasoning.

Research2026-09-12

ISACA: Point-in-Time AI Compliance Cannot Survive Legal Scrutiny

ISACA's practitioner guidance argues that legally defensible AI governance requires continuous, lifecycle-spanning evidence, not periodic sign-offs. The piece identifies a live AI inventory, named ownership, and documented legal and risk bases as the minimum conditions for defensibility. Organizations relying on static compliance documentation face significant exposure under active regulatory and litigation environments.