Algorithmic Accountability

Algorithmic Accountability in HR is the principle that employers remain legally and ethically responsible for employment decisions made or substantially influenced by automated tools – ATS AI, resume screening algorithms, video interview analysis, candidate-job matching engines. AI does not transfer discrimination liability. Also called: AI accountability, automated decision accountability, AEDT compliance.

The core principle: AI does not transfer liability

The defining doctrine of algorithmic accountability is that employers cannot outsource discrimination liability to AI vendors. The EEOC’s 2022 and 2023 guidance documents are explicit: under both Title VII and the ADEA, an employer will be held liable for the actions or inactions of an outside vendor who designs or administers an algorithmic decision-making tool on the employer’s behalf, and cannot rely on the vendor’s assessment of the tool’s disparate impact.

The legal mechanism is straightforward. If an AI screening tool produces disparate impact against a protected class – older workers, women, racial minorities, applicants with disabilities – the employer is liable for that disparate impact under existing anti-discrimination law, regardless of whether the employer built the tool or licensed it.

This has direct procurement implications. Employers must conduct due diligence on AI hiring tools, document the bias audits, retain bias-mitigation evidence, and treat AI vendors as agents of the employer for discrimination purposes.

Mobley v. Workday: the landmark case

*Mobley v. Workday* is the most-watched AI-hiring class action and the practical anchor for algorithmic accountability doctrine:

Derek Mobley, a Black male over 40 with a disability, alleged that he applied for over 100 positions through employers using Workday’s AI-based applicant tracking system and was systematically rejected, often within minutes of submission, based on what he alleged were biased screening algorithms. He sued Workday directly, alleging that Workday’s AI tools functioned as an agent of the employers using them and therefore caused unlawful discrimination under Title VII, the ADEA, and the ADA.

The federal district court ruled that Workday could be considered an “agent” of the employer for purposes of anti-discrimination law, allowing the case to proceed. The case was expanded to cover HiredScore AI and certified as a class action. Whether or not Mobley ultimately prevails, the doctrinal point is established: AI vendors providing employment-decision tools can be liable as agents of employers, and employers using those tools remain liable for the discriminatory outcomes they produce.

The federal regulatory architecture

• EEOC Strategic Enforcement Plan 2023-2027. Explicitly identifies AI-driven employment decisions as a priority enforcement area. Existing anti-discrimination statutes apply to AI-driven decisions without modification.
• EEOC May 2022 Guidance on the ADA. Addresses how AI assessments may improperly screen out applicants with disabilities.
• EEOC May 2023 Guidance on Title VII. Confirms that the four-fifths rule and traditional disparate impact analysis apply to AI-driven screening.
• DOJ joint statement (2023). Issued with EEOC, CFPB, and FTC, affirming that existing federal civil rights laws apply to automated systems.

State and local laws on algorithmic accountability

NYC Local Law 144 requires an annual independent bias audit of any automated employment decision tool used for screening, employment decisions, or promotion decisions. The audit must be publicly posted on the employer’s website, and candidates must receive at least 10 business days’ notice before the AEDT is used.

Global frameworks: EU AI Act and beyond

The EU AI Act is the most consequential algorithmic accountability framework outside the US. Key elements:

• Risk classification. Employment-related AI – recruitment screening, performance evaluation, work allocation – is classified as “high-risk,” triggering substantial compliance obligations.
• Conformity assessment. High-risk AI systems require pre-market conformity assessment, ongoing monitoring, and registration in an EU database.
• Transparency obligations. Meaningful information about the AI’s logic must be provided to affected individuals.
• Human oversight. Natural persons must be able to oversee high-risk AI, intervene, and override decisions.
• Enforcement. Fines up to €35 million or 7% of global annual turnover for the most serious violations.

Building an algorithmic accountability program

• Inventory. Catalog every AI and automated decision tool used in hiring, performance management, scheduling, and termination.
• Procurement diligence. Standard requirements for AI tools: validated bias audit results, training data documentation, model architecture overview, explainability features, NYC LL 144 compliance documentation where applicable.
• Validation against EEOC Uniform Guidelines. Selection procedures must be validated under the 1978 EEOC Uniform Guidelines. Validated job-related skills assessments meet this requirement; un-validated AI scoring typically does not.
• Notice to candidates. Where required by NYC LL 144, Illinois Video Interview Act, Maryland HB 1202, or other applicable law, provide advance notice that AI is being used.
• Bias audit and ongoing monitoring. Annual independent bias audit where required; quarterly internal monitoring of funnel conversion rates by protected class.
• Human oversight protocols. Define when AI decisions are advisory vs determinative.
• Documentation retention. Bias audit results, vendor due diligence, candidate notices, and adverse-impact analyses retained for the statute of limitations period.

See also algorithmic transparency for the candidate-facing complement to this employer-side responsibility framework.

Frequently asked questions