Guide · Published November 14, 2026 · Reviewed November 14, 2026 · Ingenious AI team · Not legal advice
ADMT Notices for Lending and Credit Decisions: What Lenders Must Tell Applicants
The short answer: Credit is the decision type every regime in the kit's data names. Colorado covers lending and credit as a consequential decision: pre-use notice, 30-day adverse-outcome explanation, correction, human review, 3-year records. California covers the provision or denial of financial or lending services as a significant decision: risk assessment first, pre-use notice, opt-out where given, access answers on logic, outcome, and purpose. Connecticut's profiling rights already apply to financial and lending effects. And creditworthiness evaluation is on the EU AI Act's Annex III high-risk list, with deployer duties from December 2, 2027.
Check your duties and dates — free Build your first notice — free
What is a lending decision here? In the kit's rules data, lending and credit covers loans and underwriting support (Colorado's decision type). California's list speaks of financial or lending services; Connecticut's, of financial or lending services among its significant effects; the EU's Annex III names creditworthiness evaluation. If a tool scores, ranks, or prices a borrower, the duties in this guide are the starting point — confirm your combination in the free checker.
Colorado credit: the explanation is the document applicants judge you by
A denied loan or a reduced line is an adverse outcome, and Colorado's sequence applies in full: pre-use notice before the tool runs, naming the tool and the lending decision; plain-language explanation within 30 days stating the principal reasons and the tool's role; correction and human review paths on the document; records for 3 years. Credit files are data-dense, so the reasons discipline matters: name the actual principal reasons from the decision record, not the category labels the tool displays.
Lenders sometimes ask whether an existing adverse-action practice already covers this. This guide takes no position on other regimes — the kit covers the SB 26-189 duties as its data states them, and its record is built to sit alongside, not replace, whatever else a lender keeps. What the kit adds is the per-decision chain: notice, reasons, data summary, explanation date, counted against the 30 days. The clock mechanics are in Colorado's 30-Day Explanation Clock.
California credit: assessment, opt-out, and the security exception's boundary
California's order of operations for a lender: complete the risk assessment before use (purpose, data involved, risks, safeguards — credit data provenance belongs in the data element); give the pre-use notice naming opt-out and access rights; offer two easy opt-out routes where the opt-out applies; answer access requests with the logic, outcome, and purpose. A business starting lending ADMT after January 1, 2027 complies from its first use.
Watch the security exception's edge. A fraud tool used solely for fraud prevention may sit outside the opt-out — but the moment its score also influences the credit decision, solely ends. Lenders run fraud and credit tooling in the same pipeline more often than any other sector, which makes the output map (which score feeds which decision) a document worth having before an examiner or a consumer asks. See California ADMT Opt-Out Requirements.
Worked example: Cascade Community Credit Union
Example — fictional business, real duty structure. Cascade is a fictional credit union using an underwriting-support tool for Colorado and California applicants. One month under the duties:
- Applications processed: 420 applications, each preceded by a pre-use notice naming the underwriting tool and the lending decision.
- Colorado explanations: 96 adverse outcomes, all explained within 30 days (median day 7), each stating principal reasons drawn from the decision record.
- California: risk assessment refreshed before the month (data element now lists 11 input categories with sources); 13 opt-outs honored via manual underwriting; 8 access requests answered with logic, outcome, and purpose.
- Exception boundary checked: the fraud tool's output map is re-verified — its score feeds no credit decision, keeping the security-only position documented rather than assumed.
- Record kept: 420 notices, 96 explanations, 13 opt-outs, and 8 access answers in one chain, held for three years; the EU December 2, 2027 date sits in the same tracker for its EU-facing pilot.
Connecticut credit: profiling rights, already live
Connecticut needs no 2027 wait for credit: financial or lending services are on the CTDPA's list of legal or similarly significant effects, so the profiling opt-out (answered within 45 days, one extension allowed, appeal on denial), the outcome-question rights (question the result, learn the reasoning, review the data used), and the data-protection assessment before heightened-risk profiling all already apply to a covered controller.
Since July 1, 2026 the opt-out reaches decisions even where a person is nominally involved, because 'solely' was removed. A lender whose workflow is 'the tool recommends, a person signs' should not assume the human signature takes the decision outside Connecticut's rights. The detail is in Connecticut Profiling Opt-Out Rights.
The EU clock lenders should already have diarised
Creditworthiness evaluation is expressly on the AI Act's Annex III list in the kit's checker data, alongside employment and education access. The high-risk deployer duties — use the system as instructed, keep human oversight, keep the logs — were postponed by the Digital Omnibus (Regulation (EU) 2026/1744) and apply from December 2, 2027. Article 50's transparency duties already apply since August 2, 2026.
A lender with EU exposure therefore runs two 2027 clocks: the US states' January 1, 2027 and the EU's December 2, 2027. The kit's record covers the US duties; its deadline tracking keeps the EU date visible alongside, which is the practical reason to keep one tracker rather than per-region lists that drift. See EU AI Act High-Risk Decision Systems.
Reasons, data summaries, and the record a lender can defend
Across all four regimes the same three artefacts recur: a notice given before the tool ran, a data summary showing what fed the decision, and reasons stated in plain language. Lenders who capture all three at decision time can answer a Colorado explanation, a California access request, and a Connecticut outcome question from the same file. Lenders who capture none are reconstructing — slowly, and under a clock in Colorado's case.
The kit's contribution is deliberately narrow: it generates the notices and explanations, runs the review queue, and keeps the hash-chained record for the retention duty. It never judges whether a lending decision was fair, and its documents say so. Build your first notice in the free generator and check your duty combination in the free checker before the next credit run.
- Map which score feeds which decision — especially fraud versus credit — and keep the output map on file
- Colorado: run the notice → 30-day explanation sequence with reasons captured at decision time
- California: refresh the risk assessment's data element until it names every input category and source
- Connecticut: treat the profiling opt-out and outcome questions as live duties today, with the 45-day answer standard
- Diarise December 2, 2027 for creditworthiness if you have EU exposure — Annex III deployer duties land then
Sources, as carried in the kit's own data: the duty summaries in app/rules.py and the checker data in app/checker_data.py, verified against primary or authoritative sources on October 28, 2026. Run the free duties checker for your jurisdiction, decision type, and the tool's role, and compare the kit's plans for the record, builder, and review queue. More from Ingenious AI is on the hub.
Related guides
- Colorado's 30-Day Explanation Clock: How It Runs and How Not to Miss It
- California ADMT Opt-Out Requirements: Two Easy Ways, Three Exceptions
- Connecticut Profiling Opt-Out Rights: What the CTDPA Gives Consumers and What Businesses Must Do
- EU AI Act High-Risk Decision Systems: The December 2, 2027 Deployer Duties
See all ADMT guides for the full set.
Frequently asked questions
Is credit scoring covered by ADMT rules?
Yes, in every regime in the kit's data: lending and credit (Colorado), financial or lending services (California and Connecticut's lists), and creditworthiness evaluation on the EU's Annex III list.
What must a Colorado lender send after denying a loan?
A plain-language explanation of the principal reasons and the tool's role within 30 days of the decision, with correction and human-review paths named.
Does a fraud tool remove the California opt-out duty?
Only if the tool is used solely for security, fraud prevention, or safety. If its output also influences the credit decision, the exception no longer describes the use.
When do EU high-risk duties for creditworthiness start?
December 2, 2027, after postponement by the Digital Omnibus (Regulation (EU) 2026/1744), per the kit's checker data. Article 50 transparency duties already apply since August 2, 2026.
How long must lending decision records be kept?
Colorado requires compliance records for at least 3 years; the kit keeps one record for all states, held for three years.