---
id: "concept-correlated-ai-errors"
type: "concept"
source_timestamps: ["§ The Business Consequences of Non-Diversity in Agentic Teams"]
tags: ["systemic-risk", "risk-management", "vendor-lock-in"]
related: ["concept-structural-ai-diversity", "claim-uniformity-compresses-differentiation", "concept-model-portfolio-governance"]
definition: "Systemic vulnerabilities created when multiple actors rely on the same AI models, causing them to experience identical failures or blind spots simultaneously."
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-new-28-agent-teams-different-models"
sourceUrl: "https://hbr.org/2026/06/the-strongest-teams-of-ai-agents-will-be-built-using-different-models"
sourceTitle: "The Strongest Teams of AI Agents Will Be Built Using Different Models"
---
# Correlated AI Errors

Correlated AI errors are a systemic risk that arises when an entire industry — or an entire organization — relies on the *same* underlying foundation models. Because such models share training data, architectures, and blind spots, they tend to fail in the exact same way under the same conditions.

The article's flagship example: in regulated industries such as **payments** or **insurance**, if all firms run the same AI stack, the entire sector might experience the same **fraud false negatives simultaneously**. This transforms what would normally be an isolated vendor risk into a massive, **industry-wide systemic vulnerability**.

The structural remedy is [[concept-structural-ai-diversity]] (uncorrelated models fail differently); the governance remedy is [[concept-model-portfolio-governance]]. Correlated errors are also the mechanism behind competitive convergence — see [[claim-uniformity-compresses-differentiation]].

**Enrichment nuance:** The *risk logic* (shared models ⇒ shared blind spots ⇒ correlated failures ⇒ systemic risk) is well aligned with AI-governance and systemic-risk literature — financial regulators warn of ML **common-mode failures** analogous to correlated credit risk, and PwC stresses modular, system-level validation because failures propagate when architectures are similar. The specific *industry-wide fraud false-negative scenario*, however, is a **hypothetical illustration**, not a documented event.


## Related across articles
- [[concept-paradox-of-access]]