HighStereotyping●Realized

Representational harm & stereotyping

Description

Outputs reinforce demeaning stereotypes or representational harms about groups.

Example scenario

Marketing copy generated for a product reflects gender stereotypes about who manages money.

Real-world evidence●Realized

Regulators have confirmed discriminatory outcomes from algorithmic fraud and credit systems in production. The Apple Card/Goldman Sachs investigation by the New York DFS found the credit algorithm assigned lower limits to women than to similarly qualified men, and algorithmic lending bias has triggered CFPB enforcement actions.

Apple Card Algorithm Investigated for Gender DiscriminationNew York Department of Financial Services / Bloomberg · 2019 CFPB Supervisory Highlights: Fair Lending Violations in Automated UnderwritingConsumer Financial Protection Bureau · 2022

Primary mitigations

Toxicity & stereotype classifiers
red-teaming for representational harm
safe-completion templates
reviewer guidelines.

Detection signals

Stereotype-benchmark scoring; harmful-association detection; sampled human review.

Mitigating controls

Non-agentic controls

ZYC-FAIR-001Non-Agentic

Group Fairness

Fairness

ZYC-ETH-001Non-Agentic

Non-Maleficence

Ethical Governance

ZYC-ACCU-001Non-Agentic

Lexical Similarity

Accuracy

ZYC-FAIR-012Non-Agentic

Intersectional Parity

Fairness

ZYC-FAIR-002Non-Agentic

Individual Fairness

Fairness

ZYC-FAIR-003Non-Agentic

Stereotype Bias

Fairness

ZYC-FAIR-004Non-Agentic

Representation Bias

Fairness

Risk scoring

Severity

High

score 3

Likelihood

High

score 3

Risk

High

Control depth

7 mapped (above avg 5.2)

Framework refs

5/8 taxonomies

Framework mappings

Regulatory4/5

EU AI Act: Annex III point 5(b) (high-risk AI system for access to financial services/creditworthiness decisions affecting individuals); Art. 9(2) (risk management system must identify and mitigate discriminatory outcomes as reasonably foreseeable risks); Art. 10(2)(f) (training data must be examined for possible biases that may lead to prohibited discrimination); Art. 10(5) (processing of special categories of personal data permitted to detect and correct bias); Art. 13 (transparency obligations — provider must disclose intended purpose and limitations relevant to affected persons); Art. 14 (human oversight measures to detect and address discriminatory outputs)
DORA: — Not covered by DORA — discriminatory outcomes in fraud detection are a fairness and ethics matter; DORA Art. 5–6 govern ICT governance and risk management frameworks but do not address demographic bias or differential treatment in model outputs
RBI: Fairness and Equity
NIST AI RMF: MEASURE 2.11
ISO/IEC 42001: Clause 6.2

Risk taxonomies3/4

MIT AI Risk Repository: Domain 1 – Discrimination & toxicity
IBM AI Risk Atlas: Toxicity / fairness
NIST ARIA: Harmful Bias & Homogenization
MAESTRO: — Not mapped

Security2/4

OWASP: — Not covered by OWASP LLM 2025
MITRE ATLAS: — Not covered by MITRE ATLAS
Cisco AI Risk: Safety taxonomy: Hate & harassment
Palo Alto AI Risk: Prisma AIRS: Toxic Content

Capability‑pattern applicability

Classification

GenAI‑specific

Capability pattern

Content-generation

Pattern scope

Content

Why

Generative-AI risk realised without agency/tool-use — applies to chat, RAG, summarization, classification, and content-generation deployments.

Ownership & threshold

Owner (RACI)

Responsible AI / Fairness Lead [A]

Threshold

Fairness metric within ±tolerance; disparity alert on breach

Inherent vs residual

Inherent

High

Residual

5.4

Medium

Mapped controls: 13
Assumed control effectiveness: 40%

Modeled, not measured — design-stage; controls not yet tested (reduction capped 40%). Residual applies a uniform assumed reduction to the inherent score; it is not evidence of tested control performance.

Related risks in Fairness & Bias

ZNR-FB-001High

Discriminatory output

ZNR-FB-003Medium

Output homogenization / monoculture

ZNR-FB-004High

Disparate quality across languages/dialects

ZNR-FB-005Medium

Proxy / indirect discrimination via prompt features

Search the Atlas

Representational harm & stereotyping

Description

Primary mitigations

Detection signals

Mitigating controls

Related risks in Fairness & Bias