---
type: "synthesis"
tags: ["synthesis", "shadow-ai", "knowledge-quality", "emergent"]
sources: ["execution"]
id: "cross-shadow-ai-fuels-decay"
sourceVaultSlug: "hbr-seg-execution"
originDay: 8
articleStem: "hbr-seg-execution"
sourceUrl: "(unified vault: 7 sources)"
sourceTitle: "HBR — Firm Ⅱ-C · Execution quality — correct execution of AI"
---
## A vicious loop no single article names

A076 and A054 never cite each other, but stacked together they describe a self-reinforcing loop that is invisible from inside either article.

1. **Low trust → hiding.** [[claim-trust-predicts-hiding]] and the [[concept-efficiency-tax]] drive employees to conceal AI workflows ([[concept-ai-knowledge-hiding]], [[concept-suppression-of-solutions]]).
2. **Hiding → no quality control.** If AI use is hidden, no one can verify it — which is precisely the condition A054 says produces [[concept-workslop-d8]] and [[concept-knowledge-decay]]. Hidden AI output enters processes unvetted.
3. **Unvetted output → verification burden.** [[claim-verification-negates-productivity]] — colleagues downstream must disentangle signal from hallucination, if they even know AI was used.
4. **Governance response → more hiding.** A054's instinct is [[action-track-provenance|provenance tracking]] and [[action-restrict-unstructured-inputs|restriction]]; but A076's [[contrarian-governance-increases-hiding]] shows logging *increases* hiding in low-trust settings — deepening step 1.

## The emergent insight

The two articles' prescriptions are in latent conflict (see [[cross-governance-vs-psychological-safety]]): A054 wants to *see and control* AI use; A076 warns that visibility-seeking *drives it underground*. The only escape from the loop is to make disclosure safe **before** demanding provenance — solve trust (A076's five commitments) so that the provenance and verification discipline (A054) has honest inputs to work with. Governance without trust doesn't reduce slop; it hides it. This is the corpus's most important non-obvious systems interaction.