---
id: "concept-attribution-uncertainty"
type: "concept"
source_timestamps: ["§ Where Trust Breaks Down"]
source_url: "https://hbr.org/2026/02/how-to-foster-psychological-safety-when-ai-erodes-trust-on-your-team"
source_title: "How to Foster Psychological Safety When AI Erodes Trust on Your Team"
tags: ["black-box", "error-resolution", "sense-making"]
related: ["claim-ai-errors-ripple-differently", "concept-trust-ambiguity", "prereq-collective-sense-making"]
definition: "The inability to trace or understand the root cause of an AI error due to its black-box nature, preventing teams from preventing its recurrence."
sources: ["adoption"]
sourceVaultSlug: "hbr-seg-adoption"
originDay: 9
articleStem: "hbr-cl-79-psychological-safety-ai-trust"
sourceUrl: "https://hbr.org/2026/02/how-to-foster-psychological-safety-when-ai-erodes-trust-on-your-team"
sourceTitle: "How to Foster Psychological Safety When AI Erodes Trust on Your Team"
---
# Attribution Uncertainty

**Attribution uncertainty** arises when a team knows *something went wrong* because of an AI output but has **no clear pathway to understand the root cause**. Because of the "black box" nature of generative AI, teams cannot check the methodology, challenge the assumptions, or reconstruct the chain of reasoning the AI used.

Contrast this with human error. When a colleague errs, the team can ask contextual questions — *"Were you rushing? What data did you use? What were you assuming?"* — and from the answers build preventative steps. This is [[prereq-collective-sense-making]] in action. AI errors leave teams **without any mechanism to attribute the failure or prevent its recurrence**, which is exactly why [[claim-ai-errors-ripple-differently|AI errors ripple through teams differently than human mistakes]].

The uncertainty **compounds**: unable to explain one failure, teams begin to question *all* other AI outputs and lose their calibration of when to trust the tool — feeding directly into [[concept-trust-ambiguity]]. See also [[quote-black-box-sense-making]].

**External grounding:** The black-box / explainability problem is broadly recognized in AI-ethics and XAI literature; "attribution uncertainty" is an interpretive construct the authors introduce over it. Note a counter-perspective: interpretable models, post-hoc explanation tools, and audit mechanisms are actively maturing, which may make error attribution — and therefore collective sense-making — more feasible than the article's stronger "impossibility" framing implies.