Housing legal drafts an AI eligibility-tool contract. How should liability for discriminatory outcomes be addressed?

Housing legal must allocate liability for discriminatory outcomes between municipal policy choices and vendor model defects rather than leaving fairness harm ownerless. City decisions on eligibility rules, thresholds, override practices, and deployment scope create policy risk the municipality should own. Vendor defects—biased training data, defective model behavior, inadequate testing, or misrepresented capabilities—belong in the vendor’s risk share. Documented allocation and risk-sharing clauses make indemnity, remediation, and insurance meaningful after a complaint or disparate-impact finding, and they force both parties to engineer controls before go-live instead of arguing after harm.

Assigning all discrimination risk solely to residents who receive adverse decisions is inequitable and indefensible: residents did not design the tool or the policy. Leaving liability unspoken so neither party documents accountability invites disputes, weakens insurance coverage, and fails governance expectations for civic eligibility systems. Making the vendor immune from any model-defect claims by default shifts unmanageable residual risk onto the city and removes the vendor’s incentive to fix defective models.

Exam caveat: allocation is shared and fact-specific—not “city always pays” or “vendor always pays.” Distinguish policy design choices from technical model defects when reading similar stems.

Operational check: in the eligibility-tool contract, define discriminatory-outcome scenarios, map city vs. vendor responsibility, require bias testing and audit cooperation, set notice and cure paths, and confirm housing legal and procurement sign the risk-allocation schedule before award.