# PIR-2026-0008 - Mobley v. Workday: AI screening vendor held potentially liable as the employer's "agent" - `id`: PIR-2026-0008 - `title`: Court rules Workday's AI applicant-screening platform can be liable as an "agent" of client employers; ADEA collective conditionally certified - `date_occurred`: 2020s (Mobley alleges rejection from 100+ jobs via Workday-powered portals; covered period spans years) - `date_detected`: unknown (pattern surfaced through the plaintiff's own application history) - `date_disclosed`: 2023-02 (complaint filed, N.D. Cal., 3:23-cv-00770); agent-theory ruling 2024-07-12; ADEA collective conditionally certified 2025-05-16 - `status`: disputed (discrimination alleged, not adjudicated; doctrinal rulings are on the record) ### The agent - `agent_description`: Workday's AI-based applicant screening/recommendation platform, used across thousands of employers to score and reject candidates at scale, often with no meaningful human review of individual rejections. - `operator_type`: enterprise (vendor operating on behalf of thousands of client employers) - `autonomy_level`: autonomous-within-policy (alleged: algorithmic rejection without meaningful human review; some rejections within minutes, including overnight) - `model_stack`: unknown (proprietary; contested in discovery) - `harness`: Workday recruiting/HCM platform embedded in client employers' hiring portals - `failure_locus`: tool-mcp (vendor platform: the alleged defect sits in a vendor tool performing a delegated function for many operators - the failure-locus pattern the v0.1 amendment was created for) ### Authority - `authority_scope`: decision authority over hiring outcomes at fleet scale; data access (applicant PII across thousands of employers) - `funds_at_risk_usd`: unknown (no damages adjudicated; potentially one of the largest employment collectives ever) - `blast_radius`: fleet/systemic (one vendor's screening layer across thousands of employers; Workday's own filings state 1.1 billion applications were rejected through its tools in the covered period) ### The failure - `root_cause`: policy-violation - primary: alleged disparate impact by age/race/disability under ADEA/Title VII/ADA. Contributing: unknown (whether defect is model, configuration, or client-set criteria is the litigation's open question) - `mechanism`: Derek Mobley alleged rejection from 100+ positions through Workday-powered portals, sometimes within minutes of applying, attributing it to algorithmic bias. Judge Rita Lin (2024-07-12) denied dismissal on the theory that Workday acts as an "agent" of client employers when its tools perform the delegated function of screening and rejecting candidates - mapping agency law onto automated systems. On 2025-05-16 the court conditionally certified an ADEA collective; notice goes to a pool potentially including hundreds of millions of applicants. - `adversary_present`: no - `exploitation_status`: in-wild-malfunction (production screening of real applicants; no attacker; harm alleged, not adjudicated - reclassified from in-wild-exploited, which implies an adversary) ### Impact - `severity`: degraded (no adjudicated loss; systemic exposure documented via certification) - `direct_loss_usd`: unknown (no judgment) - `indirect_loss_usd`: unknown - `downtime`: n/a - `data_exposure`: none ### Detection and recovery - `detected_by`: third-party (rejected applicant inferring the pattern from his own outcomes) - `time_to_detect`: years - `time_to_recover`: ongoing (litigation active as of Aug 2026) - `remediation`: none on public record - `structural_fix`: none disclosed - `controls_that_worked`: none identified - no internal control surfaced the alleged pattern; detection came from an external victim ### Legal - `liability_holder`: contested - the live doctrinal question: Workday (vendor) as "agent" of client employers, i.e. vendor liability without the employer in the suit - `precedent_set`: 2024-07-12 agent-theory ruling (AI vendors performing delegated hiring functions can be directly liable) + 2025-05-16 conditional ADEA certification at unprecedented scale - `sealed_material`: unknown ### Evidence - `telemetry_grade`: operator-logs (vendor-held screening records, contested in discovery) - `sources`: - https://www.seyfarth.com/news-insights/mobley-v-workday-court-holds-ai-service-providers-could-be-directly-liable-for-employment-discrimination-under-agent-theory.html - https://clearinghouse.net/case/44074/ - https://www.lawandtheworkplace.com/2025/06/ai-bias-lawsuit-against-workday-reaches-next-stage-as-court-grants-conditional-certification-of-adea-claim/ - https://www.proskauer.com/blog/ai-bias-lawsuit-against-workday-reaches-next-stage-as-court-grants-conditional-certification-of-adea-claim - `independence`: court clearinghouse + multiple independent law-firm analyses - strong on rulings. - `aiid_incident_id`: 489 (https://incidentdatabase.ai/cite/489/) - cross-reference; primaries verified independently - `confidence`: high on the rulings and the 1.1B-applications figure (from Workday's own filings as reported); low-medium on ultimate harm - discrimination is alleged, not proven. ### Verification notes Ruling dates (2024-07-12, 2025-05-16), judge, and the 1.1 billion rejected-applications figure verified against court-record trackers and multiple independent legal analyses; all match intake. Blast_radius set to the v0.1 `fleet/systemic` tier: one vendor layer, thousands of downstream employers. Recorded under the v0.1 legal block. ### Corrections - 2026-08-19: Added `aiid_incident_id` cross-reference (AIID 489), matched against the AIID weekly database export (2026-08-17). A cross-reference, not a re-verification; no claim changed.