Generate the candidate and employee notices each jurisdiction requires from your hiring-tool inventory, schedule the bias audit, and keep the records the new laws expect — before the dates land.
The AI hiring and resume screening workflow in this library tells recruiters how to use AI; nothing told them what they now have to disclose. Four regimes apply or are about to. Illinois HB 3773 (Public Act 103-0804), effective January 1, 2026, requires employers to notify employees when AI is used for recruitment, hiring, promotion, discharge, discipline, or the terms, privileges, or conditions of employment; the Illinois Department of Human Rights issued draft rules in December 2025 proposing an annual notice plus notice within 30 days of adoption, with seven content elements — those rules were still not final at the time of writing, so treat the elements as the sensible template, not the law. New York City Local Law 144, enforced since July 5, 2023, requires an independent bias audit within one year of using an automated employment decision tool and notice to candidates 10 business days before use. Colorado SB 26-189, signed May 14, 2026 and operative January 1, 2027, replaced SB 24-205 and requires deployers to give clear and conspicuous notice at the point of interaction with a covered automated decision-making technology, to provide within 30 days after an adverse decision a plain-language description of the technology's role, to provide on request after an adverse outcome correction rights and an opportunity for meaningful human review and reconsideration, and to retain records for three years; the Attorney General enforces it with a 60-day notice-and-cure period, and the earlier bill's impact assessments were eliminated. California SB 947, signed September 30, 2026 and effective July 1, 2027, bars employers from relying solely on an automated decision system for discipline or termination, requires a human reviewer to independently corroborate the system's output, requires post-use notice to affected employees, and carries a $500 per violation penalty; separately, the California Privacy Protection Agency's ADMT regulations require pre-use notices, opt-out and human-review rights, and access to decision logic for existing systems by January 1, 2027. In the EU, the high-risk obligations for employment systems listed in Annex III of the AI Act apply from December 2, 2027. None of this is hard. It is a notice, an audit, a human review step, and a record — and it is far cheaper to set up from the tool inventory now than to reconstruct after a complaint.
Not just the resume screener. Conversational recruiting assistants such as Paradox, scoring or ranking inside the ATS, video interview analysis, skills assessments, scheduling bots that filter, and anything used for promotion, discipline, or termination recommendations. For each: vendor, what decision it informs, which data it uses, which states and cities the affected candidates or employees are in, and whether a human reviews its output before a decision. Keep it in Airtable; it is the source for every notice below.
Illinois covers AI used for the listed employment decisions. NYC covers automated employment decision tools that substantially assist or replace discretionary decisions for NYC candidates and employees. Colorado covers automated decision-making technology that materially influences a consequential decision — a non-de-minimis factor that affects the outcome by constraining, ranking, scoring, recommending or classifying; incidental or clerical uses are excluded — for decisions made on or after January 1, 2027. California SB 947 covers automated decision systems used in discipline or termination from July 1, 2027, and the CPPA rules cover ADMT more broadly. Have employment counsel confirm the coverage column — the definitions differ and a tool can be covered in one place and not another.
Use the Claude API with the inventory row as input and the templates in the assets as the frame, so each notice names the actual vendor, the actual decision, and the actual data categories. Produce: an Illinois employee notice (annual plus within 30 days of adopting a tool), an NYC candidate notice (at least 10 business days before use), a Colorado point-of-interaction notice and an adverse-decision explanation template (for January 1, 2027), and a California pre-use and post-use notice (for the CPPA deadline of January 1, 2027 and SB 947 on July 1, 2027). Counsel reviews the generated text once; after that, the inventory drives regeneration.
Job postings and the application flow for candidate notices; the employee handbook, an annual all-staff message, and the onboarding packet for employee notices; the point of interaction — the chatbot greeting, the application page — for Colorado. Collect acknowledgments through Fillout or Typeform with a timestamp, and store the acknowledgment reference against the person's record. A notice you cannot prove was delivered is a notice you did not give.
An independent auditor evaluates the tool for disparate impact by sex and race/ethnicity categories within one year of use, and annually after; the summary of results and the distribution date must be published on your careers site. Put the audit date, the auditor, and the publication link in the inventory. If your vendor offers an audit, confirm it covers your data and your use, not just a generic one — the obligation is the employer's.
For any decision the tool informs, a named person reviews the output and records that they did. In Colorado from January 1, 2027 the statutory duty is narrower: after an adverse outcome, a consumer may request, and you must provide, an opportunity for meaningful human review and reconsideration by a trained person with authority to override, to the extent commercially reasonable (SB 26-189 § 6-1-1705). Logging a reviewer on every decision is how you will be able to honor that request; it is good practice, not the letter of the Colorado rule. California SB 947's corroboration duty is different: it applies when you rely primarily on an ADS for discipline or termination, from July 1, 2027. When a decision goes against a candidate or employee in Colorado, the plain-language description of the tool's role goes out within 30 days; template it now and trigger it from the ATS or an n8n flow on the rejection event, with the data-correction route included.
Colorado requires three-year record retention; keep notices, acknowledgments, audit reports, human-review records, and adverse-decision explanations for at least that long everywhere, because it is simpler than per-state rules. Re-run the inventory each quarter — new tools, new states, new vendors — and watch the dates: January 1, 2027 (Colorado and the CPPA rules), July 1, 2027 (California SB 947), December 2, 2027 (EU Annex III), and whatever the Illinois final rules settle on.
Use these templates as-is or customize for your business.
tool_name | e.g. conversational recruiting assistant vendor | legal name of vendor decision_informed | recruitment | screening | interview_selection | hiring | promotion | discipline | termination | terms_conditions how_it_is_used | plain-language sentence a candidate would understand data_categories_used | resume text | application answers | assessment scores | video/audio | public profile data | internal performance data output_type | score | rank | recommendation | pass_fail | summary human_review_before_decision | yes | no (no = from 2027-07-01 California bars relying solely on the tool for discipline/termination, and requires human corroboration when the tool is the primary basis) human_reviewer_role | named role jurisdictions_affected | IL | NYC | CO | CA | EU | other illinois_covered | yes | no | counsel_review nyc_aedt | yes | no | counsel_review colorado_admt | yes | no | counsel_review california_ads_or_admt | yes | no | counsel_review eu_annex_iii | yes | no | counsel_review notice_versions | links to the generated IL / NYC / CO / CA notices and their version dates nyc_bias_audit_date | YYYY-MM-DD (within one year of first use, then annual) nyc_bias_auditor | independent auditor name nyc_summary_url | public careers-site link adverse_decision_template| link (CO: sent within 30 days) record_retention_until | first_use + 3 years minimum, rolling vendor_documentation_ref | technical documentation supplied by the developer (CO requires developers to provide it) last_reviewed | YYYY-MM-DD owner | named HR or TA lead
ILLINOIS — EMPLOYEE NOTICE (HB 3773 / P.A. 103-0804, in force since Jan 1, 2026; deliver annually and within 30 days of adopting a tool) "[Company] uses artificial intelligence tools in some employment decisions. Currently: [TOOL] from [VENDOR] is used for [DECISION], using [DATA CATEGORIES]. A [HUMAN REVIEWER ROLE] reviews its output before any decision is made. We do not use AI to [EXCLUDED USES]. Questions, requests for accommodation, or concerns about this use may be directed to [CONTACT]. This notice is provided under Illinois Public Act 103-0804." (Content elements follow the IDHR December 2025 draft rules: vendor, decisions affected, data categories, contact, accommodation. The rules were not final at the time of writing — confirm the current status with counsel.) NEW YORK CITY — CANDIDATE NOTICE (Local Law 144; at least 10 business days before use) "[Company] uses an automated employment decision tool, [TOOL], to assist in assessing candidates for this position. The tool evaluates [JOB QUALIFICATIONS OR CHARACTERISTICS] using [DATA CATEGORIES]. You may request an alternative selection process or accommodation by contacting [CONTACT]. A summary of the most recent independent bias audit of this tool is available at [URL]. Information about the data collected and our retention policy is available at [URL]." COLORADO — POINT-OF-INTERACTION NOTICE (SB 26-189; operative Jan 1, 2027) "[Company] uses an automated decision-making technology, [TOOL], as a factor in [DECISION]. It considers [DATA CATEGORIES]. If a decision is made against you, you will receive an explanation of the technology's role and may request correction of inaccurate personal data and meaningful human review. To request more information about this technology, contact [CONTACT]." (Colorado allows this to be a prominent public notice linked at the point of interaction; it does not have to be an in-chat message.) COLORADO — ADVERSE-DECISION EXPLANATION (send within 30 days of the decision) "On [DATE], [Company] made a decision regarding [POSITION / ACTION]. An automated decision-making technology, [TOOL], was a factor in this decision. In plain terms, it [WHAT IT DID — e.g. ranked applications against the posted qualifications using the information in your application]. The decision was reviewed by [ROLE]. If you believe information used was inaccurate, you may request correction and human review by contacting [CONTACT] by [DATE]." CALIFORNIA — PRE-USE NOTICE (CPPA ADMT regulations; existing systems by Jan 1, 2027) "[Company] uses automated decision-making technology, [TOOL], for [PURPOSE]. It uses [DATA CATEGORIES] and produces [OUTPUT]. You have the right to opt out and request human review, and to access information about the logic involved. To exercise these rights, contact [CONTACT]." CALIFORNIA — POST-USE NOTICE, DISCIPLINE OR TERMINATION (SB 947; effective July 1, 2027) "An automated decision system, [TOOL], was used in connection with the [DISCIPLINE / TERMINATION] decision communicated to you on [DATE]. The automated decision system was the primary basis for this decision. The system's output was independently corroborated by [NAMED HUMAN REVIEWER, ROLE] before the decision was made. You may request further information by contacting [CONTACT]. You are protected from retaliation for asking about, or exercising your rights regarding, the use of this system."
CALENDAR (as of 2026-10-03 — re-verify each quarter)
In force now:
2023-07-05 NYC Local Law 144 enforced — annual bias audit, 10-business-day candidate notice
2026-01-01 Illinois HB 3773 — employee notice for AI in employment decisions (IDHR rules not final)
Coming:
2027-01-01 Colorado SB 26-189 operative — point-of-interaction notice, 30-day adverse-decision explanation, correction rights and human review on request after an adverse outcome, 3-year records
2027-01-01 California CPPA ADMT rules — pre-use notice, opt-out / human review, decision-logic access for existing systems
2027-07-01 California SB 947 — no sole reliance on ADS for discipline/termination; human corroboration; post-use notice; $500 per violation
2027-12-02 EU AI Act Annex III — high-risk obligations apply to employment systems
QUARTERLY TASKS
[ ] Re-run the tool inventory; add new tools, states, vendors
[ ] Regenerate notices for any changed row; counsel reviews changes only
[ ] Confirm NYC audit date is within 12 months; summary published
[ ] Sample 10 human-review log entries for completeness
[ ] Confirm records older than 3 years are the only ones being purged
HUMAN-REVIEW LOG (one row per decision the tool informed)
decision_id | date | tool | candidate_or_employee_id | tool_output | reviewer_name | reviewer_role | reviewed_at | reviewer_agreed (yes/no) | override_reason | final_decision | adverse (yes/no) | adverse_explanation_sent_at (CO: within 30 days) | post_use_notice_sent_at (CA discipline/termination)
RULE: from July 1, 2027 a California discipline or termination decision that relied primarily on the tool with no reviewer_name is a violation; everywhere else, the reviewer row is what lets you honor a Colorado reconsideration request and defend the decision later.Get a new AI workflow every week. Prompts, tool stacks, and ROI math included.
AI does the categorization or first-draft work, a human approves before action is taken. The pattern of choice for anything irreversible, externally visible, or financially sensitive.
Learn the agentic glossary →Where this workflow tends to break in production — and what to put in place before you ship it.
A hiring tool is in use that nobody listed in the inventory
Mitigation: Quarterly inventory re-run; procurement requires an inventory row before any HR tool is purchased or trialed.
Notice delivered but delivery cannot be proven
Mitigation: Acknowledgments collected with timestamps through a form tool and stored against the person's record; postings archived with dates.
Colorado adverse-decision explanation misses the 30-day window
Mitigation: Explanation templated and triggered automatically from the ATS rejection event via n8n; weekly report of pending explanations.
Discipline or termination decision made on tool output with no human corroboration
Mitigation: Human-review log required for every informed decision; a decision without a named reviewer is blocked in the HR workflow.
Generated notice asserts a rule that was still in draft or not yet effective
Mitigation: Templates carry the status and date of each rule; counsel reviews once; calendar re-verified quarterly before regeneration.
This is a compliance scaffold, not legal advice: the coverage definitions differ by jurisdiction, several rules were still in draft or had not yet taken effect at the time of writing, and employment counsel must confirm the coverage column and the notice text before anything is published. Skip the NYC steps if you have no NYC candidates or employees, and skip Colorado and California rows until you hire or employ there — but do not skip the inventory, because it is what tells you when that changes. Do not rely on a vendor's generic bias audit or a vendor's notice language as if it discharged your obligation; the duties fall on the employer. And do not let the tool make or substantially drive a discipline or termination decision without a named human who actually reviewed it — that is the practice California now bars from July 1, 2027, and the one a Colorado reconsideration request or a disparate-impact claim will test everywhere else.
A phased approach to get this workflow running and delivering ROI.
Days 1–30
Foundation
Days 31–60
Optimization
Days 61–90
Scale
On 30 September 2026 the FCC voted to let a STOP sent to a reminder stop only reminders, while a STOP sent to a promotion still stops all marketing. Here is how that sorts your reminder, review-request and reactivation texts into buckets, and what to keep on file.
Both now ship AI agents that categorize and reconcile. The feature gap is smaller than the price gap, and the price gap is mostly about one thing nobody mentions: user seats.
Per-minute platform pricing makes building look cheap. It is cheap. The platform was never the expensive part — and once you see the real cost line, the build-or-buy question answers itself.
One practical AI workflow per week. No fluff.
Get the full guide with step-by-step setup, workflow templates, and copy-paste assets.