Real Safety AI Foundation / Research

Disability, Accessibility, and Civil Rights

The Underwriter's Exhibit: Employment Practices Insurance, Algorithmic Hiring, and the Disability Plaintiff's Proof

Publisher: Real Safety AI Foundation

Working paper. Not peer reviewed.

Written
July 2026
Version
v0.5
Pages
17
Download PDFPlain language slides (PDF)Export BibTeXExport RIS

Abstract

An employer runs an automated screen across thousands of applications, a disabled applicant is rejected, and she files a charge. The governance industry has a product ready for that moment: documentation. Its promise is that a well kept record, bias testing, human review, an audit trail, lets the organization show the decision was defensible. At the same moment that market matured, the insurance market reached the opposite conclusion about the same technology, and carriers began writing artificial intelligence out of the very policies that would pay an employment discrimination judgment, including, by named endorsement, the employment practices line. This Article identifies a loop between those two markets that the scholarship has not examined. Where the employment practices insurer still offers coverage of an algorithmic hiring claim rather than excluding it outright, it conditions that coverage on the production of specific artifacts, whether bias testing was performed, whether a human reviewed the output, whether the tool was validated. Those artifacts are the same ones that establish a disability plaintiff’s case and defeat the employer’s defense. The bias test that unlocks the policy is the disparate impact statistic. The human review that unlocks the policy is the adopted recommendation that carries agency liability under Mobley v. Workday, and a balanced overall outcome cannot cure a specific discriminatory barrier under Connecticut v. Teal. The validation the underwriter asks after is the business necessity showing that Albemarle Paper Co. v. Moody places on the employer and that an unexplainable model cannot supply. Where an assessment is designed to infer a mental or neurological condition, its administration before a conditional offer is a completed violation the moment it runs, as Karraker v. Rent-A-Center shows for a personality inventory. The demand for the record is not even unique to insurance: a city audit-publication mandate and the civil rights recordkeeping rules point at the same file. And the deployer’s apparent escape, running the audit under privilege, fails, because the two acts the insurance regime compels, disclosure to the underwriter and reliance on the controls as a defense, are the two classic waivers of any self evaluative privilege, the underlying facts are not privileged under Upjohn regardless, and courts have already declined to shield discrimination audits. The Article locates the remedy the litigation actually seeks, an injunction to change the design, outside insurability altogether, and argues that in this domain governance documentation does not transfer risk. It converts risk into a discoverable record while the coverage that would have paid is withdrawn.

Keywords

  • employment practices liability insurance
  • algorithmic hiring
  • automated employment decision tools
  • Americans with Disabilities Act
  • Section 504
  • disparate impact
  • business necessity
  • pre-offer medical examination
  • Albemarle Paper Co. v. Moody
  • Karraker v. Rent-A-Center
  • Mobley v. Workday
  • Connecticut v. Teal
  • Upjohn Co. v. United States
  • self critical analysis privilege
  • insurance as governance
  • discovery
  • bias audit

Plain language slides

First slide of the plain language summary of The Underwriter's Exhibit: Employment Practices Insurance, Algorithmic Hiring, and the Disability Plaintiff's ProofOpen the 22-slide summary (PDF)

Suggested citation

Gilly, Travis. "The Underwriter's Exhibit: Employment Practices Insurance, Algorithmic Hiring, and the Disability Plaintiff's Proof." Real Safety AI Foundation Working Paper, July 2026. https://realsafetyai.org/research/f3sfdf/

Other versions

This paper is also posted on SSRN.

SSRN version

References (38)

This paper cites its sources in footnotes. Each authority is listed once, where it is first cited, with its footnote number.

  1. Footnote 2.Lee Harris & Cristina Criddle, Insurers Retreat from AI Cover as Risk of Multibillion-Dollar Claims Mounts, Fin. Times (Nov. 23, 2025), https://www.ft.com/content/abfe9741-f438-4ed6-a673-075ec177dc62.
  2. Footnote 3.Daniel Bauer, Joan Schmit & Justin Sydnor, The Economics of Emerging Insurance Technologies: Theory and Early Evidence, 91 J. Risk & Ins. 809 (2024).
  3. Footnote 5.Mobley v. Workday, Inc., 740 F. Supp. 3d 796 (N.D. Cal. 2024).
  4. Footnote 6.Albemarle Paper Co. v. Moody, 422 U.S. 405 (1975).
  5. Footnote 7.29 C.F.R. §1607.4(A), (B) (Uniform Guidelines on Employee Selection Procedures)
  6. Footnote 8.Katherine B. Forrest, We Have No Idea What We Are Walking Into: AI and Ethical Considerations, 1534 Annals N.Y. Acad. Sci. 19, 21 (2024).
  7. Footnote 9.Univ. of Pa. v. EEOC, 493 U.S. 182 (1990)
  8. Footnote 9.Deel v. Bank of Am., N.A., 227 F.R.D. 456 (W.D. Va. 2005).
  9. Footnote 10.Volpe v. US Airways, Inc., 184 F.R.D. 672 (M.D. Fla. 1998) (reliance on internal investigation as a defense waives any privilege over it)
  10. Footnote 10.Lawndale Restoration Ltd. P’ship v. Acordia of Ill., Inc., 853 N.E.2d 791 (Ill. App. Ct. 2006) (voluntary disclosure of an audit report waives the statutory self evaluative privilege).
  11. Footnote 11.Upjohn Co. v. United States, 449 U.S. 383, 395–396 (1981).
  12. Footnote 12.Solon Barocas & Andrew D. Selbst, Big Data’s Disparate Impact, 104 Calif. L. Rev. 671 (2016)
  13. Footnote 12.Pauline T. Kim, Data-Driven Discrimination at Work, 58 Wm. & Mary L. Rev. 857 (2016)
  14. Footnote 12.Jeremias Adams-Prassl, Reuben Binns & Aislinn Kelly-Lyth, Directly Discriminatory Algorithms, 86 Mod. L. Rev. 144 (2023)
  15. Footnote 12.Nicole Sheard, Algorithm-Facilitated Discrimination: A Socio-Legal Study of the Use by Employers of Artificial Intelligence Hiring Systems, 52 J.L. & Soc’y 269 (2025)
  16. Footnote 12.Maarten Buyl et al., Tackling Algorithmic Disability Discrimination in the Hiring Process, in Proc. 2022 ACM Conf. on Fairness, Accountability, and Transparency 289 (2022).
  17. Footnote 15.Ins. Servs. Office, Inc., Exclusion — Generative Artificial Intelligence, Form CG 40 47 (eff. Jan. 1, 2026)
  18. Footnote 15.Exclusion — Generative Artificial Intelligence (Coverage B Only), Form CG 40 48 (eff. Jan. 1, 2026)
  19. Footnote 15.Exclusion — Generative Artificial Intelligence (Products/Completed Operations), Form CG 35 08 (eff. Jan. 1, 2026).
  20. Footnote 18.Tom Baker & Anja Shortland, The Government Behind Insurance Governance: Lessons for Ransomware, 17 Regul. & Governance 1000 (2023).
  21. Footnote 24.42 U.S.C. §12112(d)(2)(A).
  22. Footnote 25.Karraker v. Rent-A-Center, Inc., 411 F.3d 831, 834–837 (7th Cir. 2005).
  23. Footnote 26.Alexander v. Choate, 469 U.S. 287, 301 (1985) (an otherwise qualified individual with a disability must be provided meaningful access to the benefit the grantee offers, and reasonable accommodations in the program may be required to assure it).
  24. Footnote 26.42 U.S.C. §12112(b)(6)
  25. Footnote 26.29 U.S.C. §794.
  26. Footnote 27.Griggs v. Duke Power Co., 401 U.S. 424, 431–432 (1971) (the effects test)
  27. Footnote 27.29 C.F.R. pt. 1607 (Uniform Guidelines on Employee Selection Procedures).
  28. Footnote 27.42 U.S.C. §2000e-2(k).
  29. Footnote 27.Choate and the Act itself.
  30. Footnote 30.Miles Turpin et al., Language Models Don’t Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting, arXiv:2305.04388 (2023)
  31. Footnote 30.Tamera Lanham et al., Measuring Faithfulness in Chain-of-Thought Reasoning, arXiv:2307.13702 (2023).
  32. Footnote 32.Connecticut v. Teal, 457 U.S. 440, 442, 450–456 (1982) (the bottom line does not preclude a prima facie case or provide a defense; a racially balanced workforce cannot immunize an employer from liability for specific acts of discrimination).
  33. Footnote 38.Bredice v. Doctors Hospital, Inc., 50 F.R.D. 249 (D.D.C. 1970).
  34. Footnote 44.215 ILCS 5/155.35
  35. Footnote 44.853 N.E.2d 791.
  36. Footnote 45.Shauhin A. Talesh, Data Breach, Privacy, and Cyber Insurance: How Insurance Companies Act as “Compliance Managers” for Businesses, 43 Law & Soc. Inquiry 417 (2018) (insurers act as compliance managers shaping organizational compliance through policy language, pricing, and risk management services)
  37. Footnote 45.Omri Ben-Shahar & Kyle D. Logue, Outsourcing Regulation: How Insurance Reduces Moral Hazard, 111 Mich. L. Rev. 197 (2012) (insurers as private regulators of safety).
  38. Footnote 49.29 C.F.R. §1607.4(A).

All research