---
id: "concept-red-team-scrutiny"
type: "concept"
source_timestamps: ["§ Stage 3: Experimental/Prototyping Portfolio (Experiment)"]
source_url: "https://hbr.org/2026/01/manage-your-ai-investments-like-a-portfolio"
source_title: "Manage Your AI Investments Like a Portfolio"
tags: ["security", "testing", "risk-mitigation"]
related: ["framework-four-portfolio-stages", "action-conduct-red-teaming"]
definition: "Deliberate, adversarial attempts to break, trick, or misuse an AI system during the experimental stage to uncover vulnerabilities before production."
sources: ["spine"]
sourceVaultSlug: "hbr-seg-spine"
originDay: 1
articleStem: "hbr-foci-61-ai-investments-portfolio"
sourceUrl: "https://hbr.org/2026/01/manage-your-ai-investments-like-a-portfolio"
sourceTitle: "Manage Your AI Investments Like a Portfolio"
---
# Red-Team Scrutiny in AI

> **Definition:** Deliberate, adversarial attempts to break, trick, or misuse an AI system during the experimental stage to uncover vulnerabilities before production.

Red-team scrutiny is a critical component of the stage gate exiting the Experimental/Prototyping phase (Stage 3 of the [[framework-four-portfolio-stages]]). Before an AI system advances to Scale & Operate, it must undergo rigorous system testing that includes deliberate attempts to break or misuse the system.

This adversarial testing is designed to uncover edge cases, security flaws, ethical bypasses, or unexpected behaviors that could cause harm or reputational damage if deployed at scale. It ensures that the guardrails and ethical guidelines established in earlier stages actually hold up under hostile or unexpected conditions. Operationalized by [[action-conduct-red-teaming]].

**External grounding:** Consistent with emerging AI-assurance practice — red-teaming of generative models recommended by major labs and policy bodies — and with model risk management (MRM) regimes (e.g., OCC 2011-12) in banking and utilities.