No Federal AI Hiring Law, Four State Theories: What Indian GCCs Actually Need to Track

There is no federal AI hiring law in the United States in 2026. The EEOC quietly removed its AI employment guidance from eeoc.gov in January 2025, and nothing has replaced it at the federal level since. If you run TA or compliance for an Indian GCC that screens candidates into US-facing roles, or you sell screening software into US employers, “no federal law” sounds like less to worry about. It’s the opposite. It means four states wrote four different rulebooks, and none of them defer to each other.

The vacuum didn’t stay empty

Title VII of the Civil Rights Act still applies to AI hiring tools the same way it applies to any other selection procedure — that part hasn’t changed. What changed is that the EEOC’s specific technical guidance on how AI selection tools trigger disparate-impact liability came down from the agency’s own website, and federal agencies stopped publishing new interpretive guidance on the topic. Employment lawyers have been tracking what filled the space since: state legislatures, moving faster and less uniformly than a federal agency ever would have.

Four states, four legal theories, no shared standard

This is the part worth sitting with. According to legal analysis tracking the shift, the states that have moved fastest on AI hiring didn’t converge on one approach — they each built a different one:

California leans on disparate impact with vendor liability attached, meaning the company that built the screening tool can be on the hook alongside the employer that used it. Illinois also uses disparate impact, but adds a private right of action, so an individual candidate can sue directly rather than waiting on a regulator. Texas takes an intent-only standard, which is a meaningfully lower bar for an employer to clear than disparate impact. Colorado uses a reasonable-care standard with an affirmative defense, so a documented impact assessment and governance process can actually shield you if something goes wrong later.

Four different bars for the same underlying question: was this AI screening decision fair. A vendor or GCC operating candidates across all four states can’t build one compliance playbook and call it done — a control that satisfies Colorado’s reasonable-care defense doesn’t automatically satisfy Illinois’s private-right-of-action exposure, and an intent-only defense that would work in Texas is irrelevant in a disparate-impact state.

Why this lands on Indian GCC and BPO teams specifically

We’ve written before about California’s SB 947 obligations for India-based screening teams, and the pattern repeats state by state: US employment law doesn’t stop applying because the screening call happens out of a Bengaluru or Hyderabad delivery center. A GCC screening candidates into US roles, or an Indian vendor selling an AI screening product to a US customer, inherits whichever state’s theory of liability applies to that specific hire — sometimes several states at once if the employer has offices in more than one. Waiting for a single federal standard to simplify this is not a strategy; there’s no indication one is coming, and the four-state pattern is more likely to become a six- or eight-state pattern before it becomes a one-state pattern.

What actually protects you

None of the four theories punish an AI screening tool for existing. They punish the absence of a record: no documented rubric, no explanation for why a candidate was rejected, no audit trail a regulator or plaintiff’s attorney can review. Colorado’s affirmative defense is explicit about this — the paperwork is the protection. The other three states get there by a different route, but the practical answer is the same: know why every rejection happened, keep the record, and don’t treat “the AI decided” as an acceptable answer to a candidate, a regulator, or your own legal team.

For a delivery-center team, that usually means three unglamorous things get done before the four-state question even comes up. First, someone owns a written answer to “why was this candidate rejected” that doesn’t require pulling an engineer into the conversation — a tagged reason, not a raw model score. Second, that record has to survive as long as the state with the longest retention expectation requires, which means picking the strictest state’s clock and running every candidate record against it rather than maintaining four separate policies. Third, whoever signs off on the screening vendor contract should be asking which of the four theories the vendor’s own liability language is actually written to satisfy — a vendor that only mentions “compliance” in the abstract hasn’t answered the question.

The contrarian part of this: “no federal AI hiring law” reads like deregulation. It’s actually a bet that fifty different legislatures move faster and less predictably than one Congress ever could — which, on the evidence of the last eighteen months, is exactly what’s happening.

If you’re building or buying AI screening for US-facing hiring out of India, talk to us about what an audit-ready screening record actually looks like.

Source: The Federal Government Quietly Removed Its AI Hiring Guidance. Four States Are Writing Their Own — National Law Review