You are staring at a new checkbox on the application page: "I do not consent to AI-assisted review of my application." Illinois HB 3773 went live January 1, 2026, NYC Local Law 144 has been on the books since 2023, and California's ADMT rules took effect October 1, 2025. The question is whether checking the box actually helps you, or just files your resume in a queue nobody staffs.
The short answer: don't check the box, fix the resume
In almost every case, opting out of AI resume screening routes you into a slower queue that the employer is not legally required to build. NYC Local Law 144's enforcing regulation says employers must let candidates request an alternative selection process but are not required to actually provide one. Illinois HB 3773 has no opt-out mechanism at all. So the checkbox is mostly a NYC AEDT opt out, and even there the New York State Comptroller's December 2025 audit called the Department of Consumer and Worker Protection's enforcement of LL 144 "ineffective."
Here is what changes your odds instead:
- A resume that actually matches the posting's language, so the AEDT scores you into the "review" bucket.
- A clean, human-sounding narrative, because the humans who do read the manual pile are sick of LLM output.
- An application timed early in the requisition's life, before the recruiter's queue is 800 deep.
What the law actually forces, state by state
The US does not have one AI hiring law; it has a six-regime patchwork, and "opt out" means something different in each. Only NYC gives you a real request-an-alternative right; Illinois gives you a discrimination cause of action; California gives you access and correction rights; Colorado gives you nothing until 2027.
| Jurisdiction | Live date | Opt-out right | What it actually gets you |
|---|---|---|---|
| NYC Local Law 144 (AEDT) | Jul 2023 | Yes, on request | Notice required; employer not required to build an alternative |
| Illinois HB 3773 | Jan 1, 2026 | No | Notice + discrimination claim if the tool discriminates |
| California ADMT (CCPA) | Oct 1, 2025 | Access + correction | Right to know inputs, request human review post-decision |
| Colorado AI Act (SB 24-205) | Jan 1, 2027 | Not yet | Delayed by SB 25B-004, then narrowed by SB 26-189 |
| Connecticut SB 5 | Pending | Varies | Automated employment-related decision technology regime |
Anyone telling you Colorado's AI Act is currently governing your job application is wrong. Governor Polis signed SB 25B-004 on August 28, 2025, postponing the original February 1, 2026 date to June 30, 2026. Then S.B. 26-189, signed in May 2026, pushed implementation to January 1, 2027 and substantially reduced employer obligations. If you are applying in Denver today, the AEDT running against your resume is answerable to nobody at the state level.
The Illinois HB 3773 rule that nobody is patching
Illinois HB 3773's sleeper provision bans ZIP codes as proxies for protected classes. No major ATS vendor has publicly disclosed remediation, and location scoring is baked into most matching engines. If you live in a ZIP the model has learned to downrank, the law says the employer is liable; in practice, you will never know. That is the difference between a right on paper and one you can exercise: HB 3773 gives you a civil rights claim, not a checkbox.
The DeepMind leak: even the AI lab doesn't trust the AI screener
The strongest evidence that manual review is broken came from inside Google. On August 10, 2026, Bloomberg reported that Google DeepMind's AGI Safety and Alignment Team had been handing candidates a secondary form warning of "a non-trivial probability your CV will be screened out incorrectly or take too long to reach us." The doc was marked "PLEASE DO NOT SHARE THIS DOC WIDELY."
Read that again. The team building frontier AI systems built a manual bypass because they did not trust their own employer's HR filters to surface the right candidates. That is the smoking gun for anyone asking whether AI screening disclosure law is worth invoking.
Gartner, cited in HR Executive, August 2026. This is the volume that broke manual review.
But the same DeepMind form contained the trap. It warned applicants: "A real human will read these. These humans get really tired of reading LLM answers, because they all sound very samey." Translation: even if you get into the human queue, generic AI-polished prose gets you dropped a second time. You have to be in the machine-readable lane and sound like a person.
That is the exact split Refolk was built for. Refolk drafts from your own history - projects, titles, metrics you actually shipped - then tailors the language to the posting so the AEDT scores you correctly. What lands on a recruiter's desk reads like you wrote it, not like the fifth ChatGPT resume they have seen this hour.
Why the "manual pile" is structurally understaffed
The manual review queue is a dead queue because the US recruiter base does not scale to catch the overflow. In Refolk's index of professional profiles, there are 112,797 people in the US whose current title is Recruiter, Technical Recruiter, or Talent Acquisition. Roughly 40% of applicants are using AI to write their resumes. If even 5% of those tick the opt-out box on a Fortune 500 job that pulls 3,000 applications, that is 60 CVs into a queue with no dedicated staff.
| Segment | Count | Source |
|---|---|---|
| US recruiters and TA (Refolk's index) | 112,797 | Refolk's index, US TA titles |
| US recruiters mentioning "Colorado" in profile | 61 | Refolk's index, US + keyword |
| Colorado-focused share of national TA base | ~0.05% | Derived (61 / 112,797) |
| Candidates using AI in the application process | ~40% | Gartner, HR Executive Aug 2026 |
| NYC LL 144 first-violation penalty | $500 | Warden AI compliance guide |
| NYC LL 144 continuing-use penalty | $1,500/day per tool | Warden AI compliance guide |
Sixty-one Colorado-tagged recruiters against a national base of 112,797 is 0.05%. When Colorado's AI Act does take effect January 1, 2027, and every applicant to a Denver posting can request human review, the arithmetic does not work. There is no bench.
A right people are penalised for using isn't much of a right.
That is Jeffrey Pole, CEO of Warden AI, on the record: "If opting out just means a human reads it instead, high take-up either recreates the cost the tool removed or turns the opt-out into the slowest queue in the process."
Who actually runs the AEDT reading your resume
The two vendors named most often in public LL 144 disclosures are HiredScore and HireVue; Pfizer's LL 144 notice names both. Knowing the vendor changes what your resume needs to do.
- HiredScore: scores fit against the requisition and surfaces a shortlist to the recruiter. Keyword overlap with the JD and a clean work history are what move you up.
- HireVue: video and structured assessments in addition to resume parsing. If you see a HireVue link after submission, your resume already cleared a gate.
The play is the same for both: put the words the posting uses, in the sections the parser expects (Experience, Skills, Education), with metrics that match the seniority the JD describes.
What to do on the application page
Here is the decision, ranked by what actually moves your callback rate:
- Do not check the opt-out box on a role you want. In NYC it routes you nowhere the employer is required to build. In Illinois it does not exist. Everywhere else it flags you as high-friction.
- Do check it if the employer is in a jurisdiction where you plan to sue. If you have reason to believe the AEDT is discriminating against you on age, disability, or the Illinois ZIP-code proxy, the opt-out request creates a paper trail. Talk to a lawyer.
- Tailor the resume to the posting. Not the whole resume: the top third, the skills line, and the bullets under the two most relevant jobs. The AEDT reads top-down and stops early.
- Strip the LLM tells. "Leveraged," "spearheaded," "results-driven," and three-clause parallel bullets that all end in a metric. The DeepMind bypass humans see hundreds of these a week.
- Apply within 72 hours of posting. Recruiter attention on a req collapses after the first week. Fresh reqs get read; stale ones get filtered.
- Follow the recruiter on LinkedIn before you apply. A profile visit before your name lands in the ATS raises the odds someone actually opens it.
Refolk handles items 3 and 4 in one pass: the resume comes back tailored, with the AI-polish tics stripped, because it is drafted from what you actually did, not from a template that reads like every other applicant's.
The California ADMT wrinkle most people miss
California's ADMT rules under the CCPA (live October 1, 2025) do not give you a pre-decision opt-out, but they give you the right to know the inputs a decision was based on and to request human review after an adverse action. That is a different lever entirely.
If a California employer rejects you, you can submit a CCPA request asking what data the automated decision used. In practice, most applicants never ask, so the requests that do come in get real attention. It is a better path than the opt-out box because it is exercised after you already have a "no" and are not being penalized for asking.
The bigger pattern
The pattern across Colorado's delayed AI Act, Illinois HB 3773's ZIP-code ban, NYC LL 144's toothless enforcement, and DeepMind's leaked bypass form is the same: the law is trying to solve a scale problem the market created, and the fixes are procedural where the problem is arithmetic. There are 112,797 US recruiters. There are millions of applications. No checkbox changes that math.
The move that does change your outcome is the one you control: send a resume that the machine scores correctly and that a tired human, when it finally reaches one, actually wants to keep reading.
FAQ
Should I ever check the AI opt-out box?
Only if you are already planning to challenge the employer's process and you want a documented request on file. Otherwise the box routes you into a queue that NYC's own enforcing regulation does not require the employer to staff, and that Illinois HB 3773 does not create at all. For a job you want, tailor the resume instead of opting out.
Does Illinois HB 3773 actually let me sue?
Yes. It amends the Illinois Human Rights Act, so discriminatory use of AI in hiring is now a civil rights violation. You need evidence the tool discriminated against a protected class, which is hard to get without discovery, and the ZIP-code proxy ban is the newest hook. Get an employment lawyer before you file; the December 11, 2025 Trump executive order directing DOJ to challenge state AI laws means litigation on the law itself is coming.
What about the Colorado AI Act, is it live yet?
No. Governor Polis signed SB 25B-004 on August 28, 2025, delaying the original February 1, 2026 date to June 30, 2026. Then S.B. 26-189, signed in May 2026, pushed implementation to January 1, 2027 and cut back employer obligations. Anyone telling you Colorado's AI Act is currently governing your job application is wrong.
How do I make my resume sound like a person and not an LLM?
Cut the resume-verbs everyone uses (leveraged, spearheaded, orchestrated), keep specific product names and specific numbers, and let one bullet be shorter than the others. The DeepMind bypass form said the humans are "tired of reading LLM answers because they all sound very samey," so uniform polish is the tell.