A decision that software makes about a person should never be the last word. Not Final asks lawmakers and regulators for three specific changes to AI governance, uses our index as the evidence, and gives people two ways to take part: a letter to the company whose AI decided about them, and a chance to tell their story.
Colorado's Attorney General published draft rules for the state's automated decision law on 11 August 2026 and is accepting public comments ahead of a hearing on 26 October 2026; the law itself takes effect on 1 January 2027. A formal rulemaking comment is the one route where a regulator must consider what is filed and answer it on the record, so the first action of this campaign is a comment with the index as evidence, asking for the first two of the three changes below. Check the written-comment deadline on the Attorney General's site; it may fall before the hearing.
The strategic route runs through Congress. Senator Edward Markey introduced the Right to Override Act in 2025 for clinical care; our third proposal extends its principle to credit, insurance, housing, employment and benefits. On 1 September 2026 he won his party's primary with about two thirds of the vote and faces the general election in November in a state that last elected a Republican senator in 2010, so the bill's sponsor is expected to return; the letter to his office and to the sponsors of Colorado's law follows the Colorado filing.
Every automated decision law passed so far gives people some rights on paper. The index shows where they stop working. These three changes close the gaps that appear most often.
Laws that give a right to human review rarely say the reviewer must have authority to overturn the decision, or set a deadline. Cigna's medical directors signed denials at a reported 1.2 seconds each; SafeRent's landlord could not override the score. We ask that every human review carry the authority to change the outcome and a fixed time to answer.
Asking for a human path should not cost more, take longer or close the door. Refusing the automated route must not count against the person. No current law says this.
The Right to Override Act, introduced in the United States Senate in 2025, would protect a clinician who overrides AI in care. The same protection is needed in credit, insurance, housing, employment and public benefits, for the reviewer and for the person.
The first edition scores 20 decisions from public records. Every company gets its file and ten business days to reply before publication. The replies are published.
People whose lives were affected by an AI decision send the six-question letter to the company or agency. What comes back, and what does not, is evidence our index can use, with their consent.
Stories go into the library with consent and a source. Names go on the letter to lawmakers. The count is published.
A formal comment to the Colorado Attorney General on the draft ADMT rules before the 26 October hearing, and the letter to the sponsors of the Right to Override Act and of Colorado's law, with the evidence file and model language for the three asks. Dates and replies appear on this page.
The same file goes to the agencies that already enforce these rights: consumer, insurance and data protection regulators in the United States and Europe, and the Global Privacy Assembly, whose 2025 resolution describes the human review we measure.
The second edition re-scores the same systems. A company that added a human option moves up. A law that created one, as Seattle's did, shows in the numbers.
First edition, methodology, codebook, letter tool and this campaign go public.
Done on the dayColorado's draft rules for the automated decision-making law were published on 11 August 2026, with a public hearing on 26 October. We file a rulemaking comment with the first edition as evidence, asking that the rules require a human reviewer with authority to change the decision and a deadline to answer, and that refusing the automated path carry no penalty. The comment is published here when filed.
First concrete filingEvery profile receives its second independent coding and the disagreement log is posted. The first letter, with the evidence file, goes to the sponsors of the Right to Override Act and of Colorado's law. The count of names closes the day before it is sent.
Date published hereThe evidence file goes to the regulators that enforce these rights, timed to the European Platform Work Directive, which member states must transpose by 2 December 2026 and which covers automated decisions about platform workers.
Replies publishedColorado's automated decision law and California's automated decision-making rights take effect. The index re-scores the affected deployments against the new duties.
Second edition beginsSame systems, new scores, plus the records cluster: background checks, data brokers and pretrial tools.
With disagreement logA letter turns a private refusal into a record.
Most people never ask. When you send the six-question letter, the company must either answer, which tells you what you can do next, or stay silent, which is itself evidence. If you agree, we add the answer or the silence to the company's file. Enough letters about one system show a pattern that a regulator can act on.
Draft the letterA story is what lawmakers remember.
Mary Louis's letter from a property manager did more for tenant screening than any statistic. Stories in the library are short, sourced and published with consent, under your name or without it. They go with the evidence file to lawmakers and regulators, and they are what reporters ask for first.
Tell your storyMany people, one system, one case.
When the same system refuses many people the same way, the law in most places allows them to act together: a class action in the United States, a representative action in the European Union, or a group complaint to a regulator. SafeRent's score was changed by a class action of voucher holders; Uber's fine began with a complaint from more than 170 drivers. The Human Option is not a law firm and does not give legal advice. With your consent, we can pass your case to organizations and lawyers who take these matters, without payment to us.
Ask about your caseThree things, and they have all happened before.
A law can be amended: Colorado rewrote its act in 2026, Seattle wrote human review into an ordinance. A regulator can act: the Dutch authority fined Uber, the FTC ordered Evolv to stop misrepresenting its scanners. A company can change its practice: SafeRent stopped showing scores for voucher holders, Uber says it now reviews every deactivation. The index records each change in the next edition.
See the scoresSenator Edward Markey introduced a bill in 2025 to guarantee a human override of AI in clinical care. Our third ask extends that principle to credit, insurance, housing, employment and benefits. The letter, with the evidence file from the first edition, goes to his office and to the sponsors of Colorado's automated decision law when the count reaches 500 names or on 15 October 2026, whichever comes first. The date of sending and the reply, if any, will be published here. The asks are written for United States law, where the gaps are widest; the evidence comes from everywhere, because the fix has already been found elsewhere. Add your name to it.
We are funding the first two editions of the index, the methodology paper, the tools and the Not Final campaign. Foundations, newsrooms, researchers and companies willing to open their decision paths are welcome.
or email info@humanoption.org