Most conversations about artificial intelligence in hiring start with the screen: the resume ranker, the chatbot interview, the model that decides who advances. But for citizenship-status discrimination, the exposure can begin in the posting itself, before anyone applies.
Take a Virginia-based IT services company’s job ads that were written by an AI tool, telling applicants that only H-1B, OPT, or H-4 visa holders need apply. Earlier this year, the Justice Department’s Civil Rights Division announced a settlement over the ads, which it alleged had restricted jobs based on citizenship status in violation of the Immigration and Nationality Act. The company agreed to pay $9,460 in civil penalties, train its staff, revise its policies, and stop running ads that exclude protected workers.
The dollar figure is small, but not the principle. The DOJ said plainly that it doesn’t matter whether an employee, a recruiter, or an AI tool wrote the ad; the employer answers for the words.
What AI Blurs
The relevant law is 8 U.S.C. Section 1324b, the Immigration and Nationality Act’s anti-discrimination provision. It bars employers with more than three employees from discriminating in the hiring, firing, or recruitment or referral for a fee based on citizenship status, where the person affected is a “protected individual.” That group includes US citizens and nationals, certain lawful permanent residents, refugees, and asylees.
Job postings go wrong when they treat three different things as one.
Citizenship is one of several statuses that carry the right to work.
Employment authorization is the legal right to work in the US. Citizens, permanent residents, refugees, and asylees have it. Many visa holders have it too, subject to the conditions of their status and often tied to a specific employer. You can only hire people who are authorized to work, and you may say so in a posting.
Sponsorship is a separate question: Will this candidate need you to file an immigration petition to get or keep that authorization? The DOJ’s own employer guidance says you generally may ask applicants whether they will need sponsorship, and you may decline to sponsor if you apply that policy evenly — not as cover for citizenship-status or national origin discrimination.
A neutral decision not to sponsor is generally permissible. But that permission isn’t a license to screen out candidates simply because their work authorization documents carry an expiration date; the Form I-9’s anti-discrimination notice warns that refusing to hire on that basis may itself be illegal discrimination.
An AI drafting tool can reproduce problematic language and assumptions found in existing job postings. Ask it for a government contracting role, and it may insert “US citizens only,” which is lawful only if a law, regulation, executive order, or government contract requires that restriction for the position. Prompt the tool with sample postings from the IT staffing market, and it may reproduce what the IT services company posted: a preference for H-1B or OPT candidates, which shuts out citizens, green card holders, refugees, and asylees from a job they are fully authorized to fill.
Absent a legal requirement for the restriction, both postings can violate the same provision for the same reason: They exclude protected workers based on status. The DOJ’s current enforcement focus, under what it calls the Protecting US Workers Initiative, is on postings that favor visa holders over US workers. The department has announced a series of similar settlements under the initiative — several involving ads restricted to visa holders.
Potential Applicants Lost
The harm in a discriminatory ad isn’t only that a particular candidate gets rejected. It is that qualified people never apply. A permanent resident who reads “must be a US citizen” likely will close the tab. A citizen who reads “H-1B or OPT candidates only” likely will do the same.
No bias audit of a screening model would surface them because they never entered the system. If you review only the ranking algorithm, you leave the earliest and arguably widest source of exclusion unexamined.
Employer’s Responsibility
Section 1324b has no software exception. It has no recruiter exception either: A staffing agency or other entity engaged in recruitment or referral for a fee carries its own liability under the statute, and using one doesn’t remove your own risk.
The DOJ’s best-practices guidance tells employers to instruct third parties that postings requiring citizenship or permanent residence, where no law requires it, are off-limits. When a language model generates the posting, the point is simpler still: The tool has no legal duties, so the responsibility for reviewing its output stays with you.
Practical Steps
Treat every AI-generated posting as a first draft that a person reviews before it goes live. Build that review around these questions:
- Does the ad mention citizenship, residency, a visa category, or any immigration status? If so, can you point to the specific law, regulation, executive order, or government contract that requires the restriction for this position? If not, remove it.
- Does it express a preference for or against visa holders? “Sponsorship available” is a legitimate signal. “Only H-1B or OPT candidates” is a citizenship-status restriction.
- Does it describe authorization and sponsorship accurately? “Must be authorized to work in the US” and “We are unable to sponsor for this position” are generally permissible. “Must be a citizen or green card holder” is a narrower statement and often an unlawful one.
Then look upstream. Audit the prompt templates and sample postings your recruiting team or vendor feeds the tool, because the restriction frequently originates there. Put approved authorization and sponsorship language in your style guide and hand it to every agency recruiting on your behalf. Keep a record of who reviewed each posting.
If you’re reviewing your AI hiring practices, you’re right to scrutinize how candidates are screened and ranked. But the first filter is the posting. If the words decide who feels eligible to apply, those words need the same scrutiny as the algorithm behind them.
The author used AI to draft a portion of this article.
This article does not necessarily reflect the opinion of Bloomberg Industry Group Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.
Author Information
Michael Ashoori is founder and president of Ashoori Law, a US immigration law firm focused on business, investment, employment, and family immigration.
Interested in writing? Review our author guidelines and submit pitches to Insights@bloombergindustry.com.