Machines now filter resumes, score assessments, and conduct first-round interviews. What most candidates do not realise is that a real set of disclosure and opt-out rights already exists, and almost nobody uses them. Here is what applies where, and what to actually do with it.
Automated screening is no longer confined to keyword matching. Amazon launched Connect Talent in April 2026, a platform that identifies, screens, and interviews candidates, including through automated voice interviews. Employers across the market are pushing first conversations to software because human capacity ran out.
Candidates have noticed and do not like it. In Gartner research, only 26% said they trust AI to evaluate them fairly, and roughly a quarter said they trust an employer less when AI is used in the decision.
Regulation here is genuinely unsettled, and you should not plan your search around being rescued by it. But the rights that do exist are concrete, currently in force, and almost never exercised, which makes them unusually useful.
Nothing below is legal advice. It is a map of what employers are already obliged to tell you, so you can recognise when the obligation is being ignored.
Local Law 144 has applied since January 2023, with enforcement from July 2023. If an employer uses an automated employment decision tool to evaluate a candidate who lives in New York City, three obligations attach.
The tool must have been independently audited for bias within the previous twelve months.
A summary of that audit must be posted publicly on the employer website. You can go and read it.
Notice at least 10 business days before the tool is used, naming the qualifications and characteristics it assesses.
That notice must also include instructions for how to request an alternative selection process or a reasonable accommodation. In plain terms: you can ask to be assessed by a person instead.
One honest caveat. The law does not compel an employer to grant an alternative process, only to tell you how to ask. Even so, a considered request marks you as a candidate paying attention, and it is free to make.
Enforcement runs through the Department of Consumer and Worker Protection rather than private lawsuits, and there is no private right of action. A December 2025 Comptroller audit found that enforcement had been ineffective, and the department has since agreed to tighten complaint handling. Penalties run from $500 for a first violation to $1,500 per day for continuing non-compliance, so employers have a growing reason to comply.
The EU AI Act classifies AI used to advertise roles, filter applications, and evaluate candidates as high risk. Those heavier obligations, including human oversight and a route to appeal an adverse decision, were pushed back from August 2026 to 2 December 2027 by the Digital Omnibus, which entered into force on 27 July 2026.
What did not move is the transparency layer. Since 2 August 2026, you must be told when you are interacting with an AI system, including a recruitment chatbot, and when emotion recognition or biometric categorisation is being applied to you.
This is the part worth committing to memory. A tool that analyses only the content of your CV is permitted. A tool that reads your facial expressions or vocal tone during a video interview to infer stress, enthusiasm, or motivation is a prohibited practice in the EU workplace context today, independent of the 2027 timeline.
Separately, recruiters using these systems are required to have documented training in how the tool works and where it fails. If the person interviewing you cannot explain what their screening tool measures, that is a signal about the employer.
Colorado passed the first comprehensive state AI act in 2024, built on the European model of risk management duties and impact assessments. It never took effect. The start date slipped from February 2026 to June 2026, a federal constitutional challenge was filed in April 2026 with the Department of Justice intervening, and enforcement was stayed.
On 14 May 2026 the state repealed and replaced it with SB 26-189, a lighter framework built on disclosure and rights around automated decision-making technology in consequential decisions, employment among them. It takes effect 1 January 2027, is enforced by the Attorney General, and carries no private right of action.
The lesson is not that regulation is pointless. It is that the timeline is unpredictable, protections are thinner than the headlines when they land, and the enforcement is administrative rather than something you can invoke yourself. Treat your own preparation as the primary defence.
Key takeaway: The rights that exist today are mostly disclosure rights. They tell you what you are being measured on. That is worth a great deal, because a published list of assessed characteristics is effectively a study guide.
Five things you can do this week, none of which require a lawyer:
If you are applying to a large employer hiring in New York City, the audit summary is on their careers site by law. It frequently names the tool and the characteristics assessed, which tells you what the first filter is looking for.
Where an employer discloses the qualifications a tool assesses, that disclosure is more reliable than guessing at keywords. Make sure your resume evidences those specific things.
If a process is fully automated and you have grounds to prefer a human assessment, ask for an alternative selection process. Keep it short and professional. The worst realistic outcome is no.
Automated video and timed assessments can disadvantage candidates with disabilities in ways the tool never accounts for. Accommodation requests are a separate and better-established right, and asking before the assessment is far easier than appealing after it.
Every one of these systems eventually hands a shortlist to a person. Writing for the filter at the expense of the human is a bad trade, because the filter cannot hire you and the person can.
There is an obvious tension in demanding a human review your application while submitting it with a bot. If automated decisions about people are objectionable when an employer makes them, they do not become fine when a candidate does.
That symmetry is the basis of how we build. We think a person should be accountable at both ends: a human decides to send it, a human decides to reject it. We set out our side of that in Where Auto-Apply Crosses the Line.
QuickApply Pro grades your resume against a specific job description and shows you what is missing, which is the legitimate version of optimising for a screen: it finds real gaps in how you have presented real experience. It also tailors your resume and cover letter, drafts answers to application questions, and prepares you for interviews, including the increasingly common automated first round.
For more on how these systems read your resume, see Why ATS-Friendly Resumes Matter.
You cannot opt out of automated hiring, and the law is not going to arrive in time to fix your search. What you can do is read the disclosures employers already publish, treat them as a specification, ask for a human when it matters, and make sure the application that reaches the shortlist is one you would defend in person.