An AI Rejected Your Job Application. Here Is What You Can Actually Demand.

Search for any of these laws and you will find dozens of well-written articles. Nearly all of them are law-firm client alerts addressed to employers: how to audit your vendor, how to paper your process, how to reduce exposure.

Very little is written for the person who got the rejection. That asymmetry is not accidental - employers pay for compliance advice, applicants do not - but it means a set of rights that genuinely exist are almost never exercised, simply because the people holding them do not know they exist.

This guide is the other side of those alerts. It covers what an applicant or employee can actually ask for, in which jurisdictions, and in what words.

Two honest caveats before we start.

First, these laws are patchy. There is no general federal right to know that an algorithm rejected you. What you have depends on where the job is, and in many places the answer is still "nothing specific."

Second, several of the strongest provisions have not taken effect yet, and at least one was significantly weakened while being delayed. Coverage written in 2025 is now wrong on the dates. That is covered below in detail, because getting it wrong is the most common error in articles on this subject.

Illinois has the broadest protection currently operating. House Bill 3773 amended the Illinois Human Rights Act effective January 1, 2026.

What it does:

• Prohibits employers from using AI that has the effect of subjecting employees or applicants to discrimination on the basis of a protected class.
• Requires notice when AI is used in recruitment, hiring, promotion, renewal of employment, selection for training or apprenticeship, discharge, discipline, tenure, or the terms and conditions of employment.
• Specifically bans using ZIP code as a proxy for a protected class.

Two features make this stronger than it first appears.

It is built into the Human Rights Act. This is not a standalone tech statute with a novel enforcement mechanism - it sits inside the existing civil rights framework, with the existing complaint process behind it. That matters, because the machinery already works.

It reaches the whole employment lifecycle, not just hiring. Promotion, discipline, and discharge decisions are covered too. If you were passed over for promotion or selected for a layoff and suspect a scoring system was involved, this applies.

The ZIP code provision deserves particular attention. ZIP code correlates strongly with race in most American cities, so a model using it can produce racially disparate outcomes while never touching race as a variable. Illinois named this explicitly rather than leaving it to be litigated.

NYC Local Law 144 has been in effect since January 1, 2023, with enforcement from July 5, 2023. It is the longest-running law of its kind and it gives applicants the most concrete, usable rights.

If an employer uses an Automated Employment Decision Tool (AEDT) on a candidate for a job in New York City, it must:

• Have an independent bias audit of the tool conducted annually, and publish the results
• Give clear and timely notice that an AEDT will be used in your evaluation
• State which job qualifications and characteristics the tool assesses
• Inform you of your right to request an alternative process or accommodation

That last point is the one to remember. In New York City you can ask not to be assessed by the tool. Very few candidates do, largely because the notice is typically buried in an application flow that nobody reads carefully.

The published bias audit is also a genuine lever. It is a public document. You are entitled to look at it before deciding how to proceed, and it will sometimes tell you the tool scored your demographic group materially worse.

A realistic note on enforcement. A New York State Comptroller audit released in December 2025 found enforcement of Local Law 144 has been weak. The rights are real and the obligations are real, but do not assume the city is actively policing this. In practice, an applicant who raises it in writing is often the enforcement mechanism.

This is where most published coverage is now out of date, so the dates matter.

Colorado. The original Colorado AI Act was set to take effect June 30, 2026. It did not. On May 14, 2026, Governor Polis signed SB 26-189, which repealed and replaced it, and delayed the effective date to January 1, 2027.

The replacement is narrower. It moved away from the original risk-based framework, dropping the duty of care aimed at preventing algorithmic discrimination and certain reporting obligations, in favour of disclosure and transparency around covered automated decision-making technology.

What survives for individuals is still substantive. From January 1, 2027, covered employers must:

• Provide notice before using covered ADMT
• Operate an adverse action process including notice, a right to correct inaccurate data, and a right to meaningful human review
• Retain records of ADMT use for three years

"Meaningful human review" is the provision with teeth. A right to have an actual person reconsider an automated rejection is the thing most applicants want and almost never get.

California. Regulations under the CCPA covering automated decision-making technology require businesses already using ADMT to be fully compliant by January 1, 2027. Businesses adopting ADMT after that date must comply before implementation. Employers must conduct a risk assessment before using ADMT for significant decisions about employees and applicants, and carry out anti-bias testing or equivalent measures. Separate California civil rights regulations on automated decision systems in employment took effect October 1, 2025.

Why this matters now, in 2026: the three-year record retention in Colorado means decisions made close to the effective date may be documented. And employers rarely build separate processes per state - many will roll out compliance nationally, which means the practical benefit often arrives before the legal obligation does.

If your job was not in Illinois, New York City, Colorado or California, you have no specific right to be told an algorithm was involved. You are not without options, though - the general civil rights framework still applies, and it applies to algorithms.

Title VII, the ADA and the ADEA do not stop applying because a machine made the decision. An employer cannot outsource discrimination to a vendor. If a hiring process produces a disparate impact on a protected group, the employer is answerable for it whether the mechanism was a manager's judgement or a model's score.

Two patterns come up repeatedly and are worth knowing:

• Disability and automated video or game-based assessments. These can systematically disadvantage candidates with speech differences, motor impairments, autism, or visual impairments. Failing to provide a reasonable accommodation is a live ADA issue, independent of any AI-specific statute. You can request an accommodation in any state.
• Age and proxy variables. Graduation year, years of experience caps, and certain technology-familiarity signals can function as age proxies.

The EEOC is the route for federal discrimination complaints, and there are strict filing deadlines - generally 180 days from the discriminatory act, extended to 300 days where a state or local agency enforces an equivalent law. Those deadlines are short and they are enforced. If you think you have a claim, the clock started at the rejection, not at the point you worked out what happened.

Many states also have their own civil rights agency with a parallel process and sometimes a longer window.

Rights you do not exercise are not worth much. The difficulty is that a vague complaint is easy to ignore, while a specific written request creates a record.

Principles that make a request effective:

• Put it in writing, by email. A record matters more than anything else here.
• Be specific and neutral in tone. Ask about process, not motive. "Was an automated tool used" lands very differently from "your robot discriminated against me."
• Name the law if one applies. It signals you know the obligation exists and routes the email to someone who can answer.
• Ask for one thing at a time. A single clear request is answered; a list of six is escalated to counsel and slowed down.
• Keep every reply. If this becomes a complaint later, the correspondence is your evidence.

A general request, any jurisdiction:

"Thank you for letting me know. So I can understand the outcome, could you confirm whether an automated tool, algorithmic scoring, or AI-assisted screening was used at any stage in evaluating my application? If so, I would appreciate knowing what characteristics it assessed."

New York City, requesting an alternative process:

"Under NYC Local Law 144, I am requesting an alternative selection process rather than assessment by an Automated Employment Decision Tool. Please also point me to the published bias audit for the tool used for this role."

Illinois, where notice was not given:

"Under the Illinois Human Rights Act as amended by HB 3773, employers are required to notify applicants when artificial intelligence is used in hiring decisions. I did not receive such a notice. Could you confirm whether AI was used in evaluating my application?"

Requesting a disability accommodation, any jurisdiction:

"I am requesting a reasonable accommodation for the assessment stage. The automated video assessment is not an accurate measure of my ability because of a disability. I would like to request an alternative format."

Request the accommodation before taking the assessment where you can. It is far more straightforward than contesting a result afterwards.

It would be dishonest to present this as a reliable path to a reversed decision. Here is the realistic picture.

What usually happens: most employers will not respond, or will respond with something non-committal. This is the common outcome even in jurisdictions with clear obligations, and it is the main reason these rights go unexercised.

What sometimes happens: a company with real compliance processes answers properly. Larger employers with counsel involved in their hiring stack are far more likely to respond substantively, because the obligation is already documented on their side.

What rarely happens: a rejection is reversed on the spot. Colorado's "meaningful human review" right, once in effect from January 2027, is the first provision genuinely designed to produce that outcome.

Where the effort is genuinely worth it:

• You have a disability and the assessment format is the barrier. Accommodation requests have the strongest legal footing and the highest success rate of anything here.
• You see a pattern - repeated rejections at the same automated stage despite clearly meeting requirements.
• You are in NYC, where the opt-out is explicit and easy to invoke.
• You have specific evidence of a protected-class effect, which turns this into a conventional discrimination claim.

Where it is probably not worth it: a single rejection from a role you were a borderline fit for, with no accommodation issue and no pattern. Sending the general enquiry costs nothing, but calibrate your expectations.

One thing worth saying plainly: being screened out by an automated system is not evidence that anything unlawful occurred. These tools reject enormous numbers of qualified people for mundane reasons - keyword mismatches, formatting, an unparsed date. Unfair is not the same as unlawful, and most of what happens here is the former.

Exercising rights after a rejection is the weaker half of this. Reducing the chance of being wrongly filtered is the stronger half.

• Assume machine-readability is a threshold requirement. Multi-column layouts, text in images, headers and footers, and unusual section titles all cause parsing failures. A rejection caused by a parser is not a judgement about you, and it is entirely preventable.
• Mirror the posting's own vocabulary. If it says "accounts receivable" and your CV says "AR," write both. Keyword matching remains the dominant mechanism in most screening stacks.
• Keep dates unambiguous in a standard format. Date parsing failures routinely produce nonsense experience calculations.
• Request accommodations before the assessment, not after the result.
• Keep records. Save the posting, the date you applied, any notice about automated tools, and the rejection. Under Colorado's law employers must retain ADMT records for three years - you should keep your side too.
• Watch the clock. EEOC deadlines are generally 180 days, or 300 where a state or local agency enforces an equivalent law. Short, and enforced.

An AI copilot can be genuinely useful in the preparation stage - checking whether your CV is parseable, mirroring a posting's language, and drafting a clear, neutral request for information. What it cannot do is tell you whether you have a viable legal claim.

If you believe you were discriminated against, talk to an employment lawyer. Many offer free initial consultations, many work on contingency, and the filing deadlines are short enough that a delay can end a claim before it starts. Use AI to arrive at that conversation prepared - with a timeline, your documents, and clear questions - not to replace it.