The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race

Policy tool · See record

Notice and appeal rights for consequential AI decisions

What the tool does

Tell people when a consequential decision is automated and give them a meaningful route to challenge it. Depending on the law, this may include an explanation, human intervention, a chance to present their position, a regulator complaint or judicial review.

Current status

Enacted in jurisdiction-specific forms. Moldova’s law supplies a concrete automated-decision rights framework. Coverage and exceptions matter. California’s newer professional and healthcare duties are related safeguards; disclosure alone should not be presented as a new general appeal right.

Where it has been used

Moldova’s automated-decision rights

Added by

Moldova’s Law No.195/2024 provides rights concerning consequential automated decisions, including human intervention, challenge, complaints to the regulator and judicial remedies. The arrangement gives affected people institutional routes for questioning an automated decision instead of relying solely on the system’s operator.

Read more about this case →

California’s clinical AI safeguards

Added by

Official status confirms AB1979 chapter 854 and SB503 chapter 857, approved and chaptered September 30. Final enrolled text preserves licensed clinical judgment and prohibits autonomous performance of functions requiring a license, with supervised trainee and nonjudgmental documentation/communication qualifications. AB1979 brings covered health chatbots within medical-confidentiality law. SB503 assigns developers intended-use, training-data, evaluation and foreseeable-bias documentation and mitigation duties, while deployers must monitor and mitigate biased impacts. These are reasonable-effort/proportionate duties, not a promise of error-free medicine.

Read more about this case →

California’s legal-professional AI duties

Added by

SB574 chapter 858 was approved and chaptered September 30. The final enrolled act prohibits lawyers from delegating the practice of law to generative AI, requires personal verification of citations and protects nonpublic information under specified conditions. Arbitrators retain decision responsibility and disclose generated information drawn from outside the record.

Read more about this case →

China’s AI litigation guidance

Added on

China’s Supreme People’s Court issued September 7, 2026 opinions on applying existing law to AI disputes. The guidance addresses evidence collection and adverse inferences when relevant evidence is withheld. It concerns how courts handle litigation under existing authority, rather than establishing a new general statutory appeal right.

Read more about this case →

Townsite’s administrative appeal

Added on

On September 1, the Interior Board of Land Appeals, the Interior Department’s administrative appeals body, stayed the Bureau of Land Management’s June 26 authorization for Townsite Solar 2 LLC’s data center near Boulder City, Nevada. The signed order consolidated the city’s and environmental groups’ appeals and expedited consideration. The proposed project is described as 167 megawatts; it is not operating capacity.

BLM had relied on environmental analysis for a previously authorized solar-and-battery project instead of preparing a new or revised assessment for the computing facility. The panel found a persuasive likelihood that this treatment violated environmental-review requirements, and granted a stay after considering the four required factors. The ruling preserves the site while the appeal proceeds; it does not permanently cancel the project. BLM subsequently sought to set aside its approval and conduct additional review, according to September reporting. October 13 AP coverage still describes construction as stalled.

Read more about this case →

Mona residents’ permit appeal

Added on

On September 15, Mona Cares submitted an appeal asking Juab County, Utah, to overturn a conditional-use permit for a proposed campus just outside Mona. The project was approved in April and is described as roughly three million square feet. Residents say they did not understand that the earlier notices concerned a data center; county officials say they followed the law and provided notice.

The appeal also faces a timing dispute: the normal filing period is 30 days, while residents contend that minutes documenting the approval did not appear online until late August. October reporting describes continuing conflict and a September 21 commission meeting; it does not establish that the permit has been overturned. A resident’s account of a pipeline for a gas plant does not certify a final behind-the-meter design. Developer identity, approved power/water specifications and cost obligations remain to be established from the permit record.

Read more about this case →

New York City proposes access to human customer support

Added on

The October 20 proposal would require disclosure of automated customer service and a clearly identified transfer or queue for human help, with alternatives when help is unavailable or waiting would be unreasonable. This is proposed, not operating: comments close November 19 and adoption remains ahead.

Read more about this case →

Italy puts a human decision-maker—and a legal remedy—between workplace AI and covered final decisions

Added on

Italy’s Legislative Decree 179, issued October 7 and published October 8, took effect October 23. Article 41 requires employers using AI in covered final employment decisions to leave the decision to a person with effective, autonomous authority. The rule covers establishing, changing or ending employment, including discipline and performance evaluation. Workers cannot waive it by consenting. Candidate searching and screening, including exclusion from later recruitment stages, are expressly outside this particular definition; other applicable protections remain.

An affected worker can request an intelligible explanation from a person, including AI’s influence and the principal factors considered. Dismissals and other employer decisions that violate the human-decision requirement are void. This gives the requirement a concrete legal consequence.

Read more about this case →

Source snapshot: 2026-10-25. Individual event and verification dates remain as recorded above. Inclusion alone does not assign a score.