Defining Ethics
Responsible AI
Ethics is a moral framework that ensures individuals make decisions that reduce harm, provide transparency, and lead to fairness. In the context of Artificial Intelligence, ethics is what allows the successful development and deployment of systems by serving as operational guardrails that protects projects from systemic failure and regulatory non-compliance.
4 Pillars of Operational Governance
A recent Stanford study followed 3.4 million AI- evaluated job applicants and identified discriminatory outcomes consistent with the Equal Employment Opportunity Commission's (EEOC) standard for adverse impact. AI systems recommend minority groups at less than 80% the rate of majority candidates, with evidence showing that 26% of black applicants and 15% of Asian applicants were excluded from roles they would likely have been selected for if they were white with comparable qualifications.
Algorithmic Discrimination
In accordance with NIST Al RMF Measure 2.5, any deployed Al system must be demonstrated to be valid and reliable through independent testing that goes beyond the conditions under which it was originally trained. In this case, the organization should have independently validated the tool by running it against resumes of their existing employees to analyze how the model actually scored and recommended candidates before full deployment.If the Al filter suddenly rejects your top-performing internal employees during baseline testing, you immediately know the model's screening logic is deeply flawed.
Under this same Measure 2.5 control, organizations must establish a defined, regular interval such as quarterly or bi-annual reviews to continuously audit the deployed model. Setting these fixed timelines ensures the compliance team regularly checks whether the AI system remains unbiased and is still capable of clearly explaining its internal decision-making logic over time.
To address NIST Measure 2.7, the organization should implement continuous red teaming activities to verify the explainability and fairness of the third-party AI system. This involves submitting paired tester resumes with identical skills but altered demographic indicators, such as changing names or swapping a standard university for an HBCU, to evaluate how often the AI flags or rejects certain profiles as part of ongoing improvement efforts.
By embedding standardized stress testing into the deployment workflow, organizations can actively identify algorithmic flaws and drastically reduce their exposure to discriminatory practices.
At its Core, a framework such as the NIST AI RMF supports companies implementing AI governance, transforming it from a simple ethical choice into a structural business requirement that protects the organization against legal liabilities and intense reputational risk.
These companies face major legal exposure through costly class action lawsuit, independent civil lawsuits filed by rejected candidates and regulatory enforcement action by the EEOC.
Businesses that build high risk AI screening applications, as well as organizations that utilize these tools via third-party systems risk severe fines reaching 15 million Euros (Approximately $17.2 million) or 3% or total global turnover. In this specific case these penalties are the result of the companies failing to maintain an active risk management framework, failing to conduct fundamental rights impact assessments, failure to ensure human oversight capabilities, and non-compliant data governance and training data practices.
The organizations face Reputational Damage and the erosion of public trust. This causes businesses to lose out on future B2B clients and customers while driving away future top talent.
The Trustworthy Review Board: Preventive Measures
Under the NIST AI RMF Govern 1.1 Function organizations must ensure staff are trained and retrained on legal considerations that impact AI deployment, meaning they must thoroughly understand existing anti-discrimination and equal employment laws within the context of AI algorithms. This training leads to staff being properly equipped to monitor the system and identify if the AI filter begins generating racially biased or discriminatory recommendations.
During the 1940s and 1950s, the Red Scare created widespread fear of communism, leading to defamation and false accusations within a hysterical climate that destroyed reputations and careers in its aim to separate the “bad eggs” from the good ones. Today, AI systems are increasingly taking on the role once played by the House Un-American Activities Committee, and protected classes are emerging as the new “communist scapegoats.”
This raises broader concerns about how AI hiring systems behave once they are deployed at scale across multiple employers using shared vendor platforms.
Stanford also found the rejection problem to be widespread, showing that applicants who submit multiple applications to companies using the same AI hiring vendor are more likely to be rejected across all roles, rather than being assessed independently on a job-by-job basis.
In contrast, in a mixed job market, where AI was not the sole decision maker, applicants were rejected at a rate consistent with companies making independent hiring decisions. AI is effectively blacklisting candidates from the job market, similar to how Hollywood Studio in the 1950s would quietly screen out actors, directors, and writers suspected of having "Communist" sympathies. You simply cease to exist with your industry.
Non-Compliance
As of writing this case study, the AI system’s creators and the companies deploying them are in non-compliance with U.S. and EU regulatory standards.
Outlined below, the table highlights several areas of non-compliance.
How Does This Impact Businesses?
AI Blacklisting is the New McCarthyism
Around 90% of employers now rely on AI tools in hiring, most which are sourced from third-party vendors rather than created in-house. This means employers often have limited visibility into how these systems operate, unless they have developed internal auditing frameworks designed to assess AI tools and mitigate potential risks. AI tools have been found to amplify harmful biases embedded in society today intensifying their impact.
Research shows that AI interview tools can discriminate against women, minorities, and people with disabilities, partly because these groups are often under-represented in the training data.
The Problem
AI Hiring Is Quietly Blacklisting Job Seekers
Case Studies
More for You
Take this 2- minute diagnostic to find out if your organization is built on stable governance or pure luck.
Assess Your Organization