← Back

Guide · Published November 14, 2026 · Reviewed November 14, 2026 · Ingenious AI team · Not legal advice

← All guides

California ADMT Opt-Out Requirements: Two Easy Ways, Three Exceptions

The short answer: Where California's CCPA ADMT rules give an opt-out, a business must offer at least two easy ways to opt out of automated decision-making technology used for a significant decision (regulation section 7221), honor each opt-out, and record that it was honored. Three exceptions can remove the opt-out duty: you offer an appeal to a qualified human reviewer who can overturn the decision; the tool is used solely for security, fraud prevention, or safety; or, for certain hiring, admission, and work-assignment uses, you evaluated the tool for accuracy and discrimination and keep safeguards in place. The duties apply from January 1, 2027 for existing use.

Check your duties and dates — free Build your first notice — free

What is the ADMT opt-out? It is a person's right, where the California rules give one, to say no to having ADMT used for a significant decision about them — and to have that no honored and recorded. It sits inside a set of four California duties (pre-use notice, opt-out, access, risk assessment) walked through together in California CCPA ADMT Requirements.

Two easy ways — what 'easy' has to mean in practice

The regulation's floor is two routes, and the kit's checker data repeats it verbatim: give people at least two easy ways to opt out. Two routes that both fail in practice — a phone line staffed one hour a week and a postal address — satisfy the count and defeat the purpose. Easy means findable from the pre-use notice itself, usable without an account the person may not have, and completable before the tool runs.

Design the pair so they fail independently: for example, an in-flow control in the application itself plus a named email contact. If one channel breaks — a form outage, a mailbox change — the other still works, and your record can show continuous availability. The notice that names the routes is the pre-use notice; an opt-out route nobody is told about is operationally the same as no route.

Honoring the opt-out is a workflow, not a sentiment

An opt-out that arrives after the tool has already scored the person honors nothing. The honoring workflow has three steps that must happen in order: the request is captured with its date, the person's file is flagged out of the automated path before the run, and the decision is made by the human route instead. The kit's rules data compresses this into one line — honor it and record that it was honored — but the record is what proves the sequence.

Partial honoring is the common failure: the tool still generates a score 'for reference' and the human reviewer sees it. Whatever your policy, the record should be able to show which files the tool touched and which it did not. That distinction is also what logging opt-outs properly is for.

Worked example: Golden Gate Home Loans

Example — fictional business, real duty structure. Golden Gate Home Loans is a fictional California lender using ADMT to support lending decisions — a significant decision under the regulations. Its first opt-out month, January 2027, looks like this:

Exception one: the qualified human-reviewer appeal

You may skip the opt-out if you offer an appeal to a qualified human reviewer who can overturn the decision. Every word carries weight: the reviewer must be qualified (able to evaluate the decision), human (not a second tool), and empowered (able to overturn, not merely to annotate). An appeal that always ends with the original outcome standing is not the exception the regulation describes.

If you rely on this exception, the appeal becomes your core consumer-facing duty, and it deserves the same design care as an opt-out: named in the notice, easy to start, decided by someone with real authority, and recorded. The kit's rules data carries the matching duty from the other side — where no opt-out applies, offer an appeal to a human reviewer for significant decisions — explored in ADMT Appeals When No Opt-Out Applies.

Exceptions two and three: security-only use, and evaluated hiring tools

The second exception is narrow by design: the tool is used solely for security, fraud prevention, or safety. Solely is the constraint — a fraud tool whose score also feeds a credit or pricing decision has left the exception behind. If you claim this exception, be able to show the tool's outputs go nowhere else.

The third exception covers certain hiring, admission, and work-assignment uses where the business evaluated the tool for accuracy and discrimination and keeps safeguards in place. Evaluation is not a one-time vendor certificate: it is your evaluation, of your use, with safeguards you maintain. Keep the evaluation with the decision record, dated, so its currency can be checked. Relying on an exception is a position you may have to explain; the file should explain it for you.

Records: the thread through all of it

California's kit duties end where the other states' do: keep records of notices and responses. For the opt-out specifically, the record set is the notice that named the routes, each request with its date, the honoring flag on the file, and — where an exception was relied on — which exception and the evidence behind it. The kit keeps one record for all states, held for three years.

Run your facts through the free duties checker before designing: it asks the jurisdiction, decision type, and the tool's role, and returns the opt-out duty in its exact regulatory wording for your combination. Then keep the design honest with the kit's record rather than a mailbox nobody audits.

  1. Confirm the opt-out applies to your use — significant decision, no exception — using the free checker
  2. Offer two genuinely easy, independent opt-out routes and name both in the pre-use notice
  3. Build the honoring sequence: capture with date, flag the file, decide by the human route — in that order
  4. If you rely on an exception, write down which one and keep its evidence (appeal design, security-only scope, or your accuracy and discrimination evaluation) on file
  5. Log every opt-out as honored in the decision record, and audit monthly that no opted-out file was touched by the tool

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.

Build your first notice — free

Related guides

See all ADMT guides for the full set.

Frequently asked questions

How many ways to opt out must a California business offer?

At least two easy ways to opt out of ADMT for a significant decision, under regulation section 7221, where the rules give an opt-out.

What are the exceptions to the California ADMT opt-out?

Three: offering an appeal to a qualified human reviewer who can overturn the decision; using the tool solely for security, fraud prevention, or safety; or, for certain hiring, admission, and work-assignment uses, having evaluated the tool for accuracy and discrimination with safeguards in place.

What happens after someone opts out?

The opt-out must be honored — the person is decided without the ADMT — and recorded as honored. The kit's rules data states the duty exactly that way.

When do California's opt-out duties start?

The compliance phase for businesses already using ADMT for a significant decision begins January 1, 2027. A business starting ADMT use after that date must comply from its first use.

Does Colorado have the same opt-out?

No. Colorado's kit duties are the pre-use notice, the 30-day explanation, data correction, human review, and 3-year records. The opt-out in this form is a California duty (and Connecticut has its own profiling opt-out).

Build your first notice — free