AI Bias Audit

State of play.

• The Trump DOJ has made Colorado's SB24-205 a federal test case for dismantling state AI bias regulation. DOJ intervened in xAI's suit, a federal magistrate issued a TRO blocking enforcement, and the Colorado AG joined the motion — leaving the nation's first comprehensive algorithmic discrimination law effectively suspended pending legislative revision or court ruling .
• The White House's March 2026 National AI Legislative Framework explicitly targets state-level bias and audit mandates as "undue burdens," proposing federal preemption and distributing oversight across FDA, FTC, DOJ, and Commerce rather than creating a new AI regulator .
• Mobley v. Workday is the live class action defining vendor liability for AI hiring tools, with ADEA claims certified and surviving dismissal while a parallel individual suit quietly settled — a combination that signals serious settlement pressure on Workday and exposure for employers using similar platforms .
• Candidate-facing disclosure failures are generating measurable litigation risk in AI hiring, with a Greenhouse survey of approximately 1,200 workers documenting that roughly 70% were not informed AI would assess them and over one-third reported age or race bias .
• For counsel advising employers, HR tech vendors, or healthcare AI deployers, the practical baseline is a bifurcated risk environment: federal preemption is moving against state audit mandates, but private litigation under existing federal discrimination statutes is accelerating simultaneously and is unaffected by that preemption push.

Where things stand.

• Colorado SB24-205 is suspended and its future turns on legislative revision. A TRO blocks all enforcement; the Colorado AG has committed not to enforce pending amendment; the legislature's window to pass successor legislation closed May 13 .
• DOJ's Equal Protection theory is the novel weapon against state bias mandates. The Trump administration's Civil Rights Division, under Harmeet K. Dhillon, argues that disparate-impact liability frameworks and diversity exemptions in SB24-205 themselves violate the Equal Protection Clause — a theory that, if adopted, would reach far beyond Colorado .
• Federal preemption architecture is being built without enacted legislation. The March 2026 National Policy Framework proposes federal preemption of conflicting state AI laws, with Commerce Department review of state measures, but no preemptive statute has passed — meaning state laws remain in force until courts or Congress act .
• Mobley v. Workday is the leading private enforcement vehicle for algorithmic hiring bias. Preliminary class certification covers ADEA claims for applicants over 40 since 2020; disparate impact and agency liability theories against the vendor have survived multiple dismissal rounds .
• AI hiring tool disclosure is an emerging standalone compliance obligation. Greenhouse survey data documents widespread nondisclosure of AI assessment to candidates, with substantial attrition and reported age and race bias — creating exposure under existing employment discrimination statutes even without new AI-specific law .
• Connecticut's proposed bias audit bill (SB00435) represents the next state-level test, requiring audits of automated decision systems with Labor Commissioner oversight — the Connecticut legislature adjourned May 6, making the bill's fate determinative for whether another state joins the audit-mandate landscape .
• AI system reliability failures are generating product liability exposure outside the hiring context. AI-powered security cameras from ADT, Ring, Avigilon, and others generate widespread false alarms despite marketed accuracy claims, with no manufacturer recalls or formal acknowledgments — a pattern that fits consumer protection and product liability theories .
• Model collapse and data provenance are emerging as long-horizon reliability and liability issues. Research drawing on Oxford and Canadian studies documents AI systems training on synthetic data and degrading over time, raising questions about professional-use liability and potential regulatory mandates for training data segregation .
• Algorithmic bias litigation now spans sectors. A Baker Donelson report documents claims centered on data privacy violations, algorithmic bias, unauthorized data use, and worker surveillance extending to organizations of every size — not just technology companies .

Latest developments.

• Federal magistrate issues TRO blocking Colorado SB24-205 enforcement on joint motion of xAI and the Colorado AG; Colorado legislature's window to pass successor legislation closed May 13 .
• DOJ formally intervenes in xAI's Colorado suit, advancing an Equal Protection theory that disparate-impact AI mandates are themselves unconstitutional .
• Greenhouse survey of approximately 1,200 workers documents 64% AI interview exposure, 38% abandonment rate, 70% nondisclosure rate, and over one-third reporting age or race bias .
• Individual Workday bias suit settles quietly while Mobley v. Workday class action advances with certified ADEA claims .
• White House March 2026 National Policy Framework proposes federal preemption of state AI laws, distributing oversight across existing agencies with no new AI regulator .
• Connecticut SB00435 bias audit bill advanced through the 2026 session with the legislature adjourning May 6 .
• AI security camera false-alarm failures documented across ADT, Ring, Avigilon, Reolink, and others — no manufacturer recalls or formal acknowledgments .
• Neuroscientist research on AI model collapse and synthetic data degradation raises professional-use liability and data provenance compliance questions .
• Baker Donelson report documents algorithmic bias litigation wave extending to all sectors, with courts establishing precedents on data ownership and corporate accountability .

Active questions and open splits.

• Will DOJ's Equal Protection theory survive judicial scrutiny — and how far does it reach? If a federal court accepts that disparate-impact AI mandates violate the Equal Protection Clause, the theory extends to any state statute requiring demographic-outcome monitoring or diversity-based exemptions, potentially invalidating a generation of state AI equity legislation .
• Does federal preemption of state AI bias mandates require enacted legislation or can executive framework documents do the work? The March 2026 National Policy Framework proposes preemption but no statute has passed; courts have not yet addressed whether Commerce Department review of state laws creates enforceable preemption .
• What is the scope of vendor liability for AI hiring tools under Mobley's agency theory? The certified class and surviving ADEA claims in Mobley v. Workday test whether an HR software vendor is an "agent" of employers for discrimination purposes — a ruling that would reshape indemnification negotiations in every AI hiring tool procurement .
• Does nondisclosure of AI assessment in hiring constitute an independent statutory violation? Regulators have not yet formally treated nondisclosure as a compliance violation under existing employment discrimination statutes, but the Greenhouse data on widespread nondisclosure and reported bias creates the factual predicate for that theory .
• Can AI product liability claims succeed on a "marketed performance vs. field reliability" gap theory? Security camera false-alarm cases would test whether the delta between marketed accuracy and actual false-alarm rates constitutes a consumer protection or product liability claim — a theory applicable across AI systems sold on performance claims .
• How will courts allocate liability between AI developers, deployers, and employers when bias is embedded in a vendor platform? The Mobley agency theory, the Colorado statute's developer/deployer distinction, and the Baker Donelson litigation wave all press toward a liability allocation framework that does not yet exist in settled doctrine .
• Will Connecticut's bias audit bill become law — and does its outcome signal the viability of state audit mandates post-Colorado? The Connecticut legislature adjourned May 6; the bill's fate is a leading indicator of whether states continue pursuing audit-mandate approaches or retreat in the face of federal constitutional challenges .

What to watch.

• Whether the Colorado federal court converts the TRO to a preliminary injunction and what constitutional grounds it emphasizes — the First Amendment compulsion theory or DOJ's Equal Protection theory will have different downstream reach for other state statutes.
• Whether Colorado's legislature passed successor legislation before the May 13 adjournment and how any revised statute addresses DOJ's objections to disparate-impact liability and diversity exemptions.
• Merits briefing and any summary judgment ruling in Mobley v. Workday on the agency liability theory — the first federal appellate ruling on vendor liability for AI hiring tools will be a market-moving event for HR tech procurement.
• Whether any federal agency (EEOC, FTC, or CFPB) issues formal guidance treating nondisclosure of AI assessment as an independent compliance obligation under existing statutes.
• Whether the Connecticut bias audit bill (SB00435) was enacted before the May 6 adjournment, and whether other states with pending AI audit legislation recalibrate in response to the Colorado litigation.
• Whether the White House's preemption framework advances to enacted legislation — the America AI Act or any successor bill that includes an express preemption clause would fundamentally reset the state-law compliance landscape.