Give affected people recourse.
Give people a way to challenge a decision about them, and to get harm put right.
Prevents A wrong decision that the person it harms can neither see, question nor reverse.
Proposed
What it asks
People should be able to discover when AI materially influences a consequential decision about them, understand enough of its basis to contest it, and reach someone with the authority to correct it.
In the hiring example, an applicant needs a route to correct an erroneous record or challenge an unsuitable assessment. Sending them a technical explanation of a model does little if nobody can reconsider the decision. A meaningful appeal needs an owner, a timescale, and the power to provide a remedy.
This also applies to people whose data enters a system. Providers should explain the uses they make of it and offer workable processes for access, correction, and deletion where applicable. They should describe technical limitations honestly, including the difference between removing a source record and undoing its influence on an already trained model.
Human agency extends beyond individual appeals. Workers, affected communities, and countries using systems developed elsewhere should have a role in shaping the standards. Access to that discussion should not depend on owning a training cluster. Participation will take funding, translation, and technical support if it is to mean anything beyond an invitation to comment.
Proposed
What compliance looks like
- Notice that AI materially influenced the decision, and understandable reasons.
- An appeal with an owner, a deadline and the authority to correct harm.
- Public participation before new uses are authorised.
Illustrative
The case that tests it
A stolen car is found. Who else is being watched?
An investigator is looking for a stolen car. A camera can help find it. Connect enough cameras, and the same tool can follow the daily movements of people who are not suspected of anything.
01Who decides what gets collected?
A camera records a vehicle passing a street. Investigators look for a car connected to a reported theft.
02Who can search across the network?
Records from many cameras become searchable together. A local tool can reveal movement across a much larger area.
03What turns a result into evidence?
A search result enters an investigation. Someone must decide how much weight it deserves before taking action.
Our inferenceA safeguard has to govern how a result becomes an action. Better matching alone cannot decide whether a use is justified.
Reported
What the case draws on
Provider account
Useful does not settle legitimate.
Flock offers automated licence plate recognition and a national network of plate readers. Its trust materials describe its approach to privacy and customer controls. Those are the provider’s account of its safeguards, rather than independent proof of how every deployment operates.
The scenario here is a general policy comparison, not a claim that every Flock customer lacks these checks. Plate recognition and facial recognition are different technologies. Both raise questions about who can search, for which purposes, and with what recourse.
Legal representatives’ account
A documented failure.
- Happened
- 2020
Robert Williams was wrongfully arrested in Detroit in 2020 after police relied on an incorrect facial-recognition result. The 2024 settlement introduced requirements for independent supporting evidence and an audit of past cases. The account and settlement links come from his legal representatives.
Reported
On the record
- Dispute-process update
Authors contest who gets paid.
Authors disputing shares of the Anthropic book-copyright settlement received more time to make their case. The Authors Guild reported an extension from 30 to 60 days, after notices exposed conflicting claims from authors, publishers and agents.
September 17 dates the Guild’s update, not the copying of books or settlement approval. The administrator confirms the extension. Disputed payments remain held back; extra time is not a resolved claim.
- Records disclosure announced
- Patients’ reports
Patients wait in pain for approval.
After suing for records, EFF reported that clinicians described patients waiting in pain and cancelled procedures in Medicare’s AI-assisted WISeR authorization program.
September 8 dates EFF’s disclosure announcement; the patient reports are from March 2026. These are clinicians’ accounts, not individual court findings. CMS requires clinical review of denials; delays span the whole process, not just an AI model.
Proposed
Where it can fail
- A review process is hollow if the reviewer cannot change the outcome.
- Governments must face scrutiny of their own deployments.
- Researchers and maintainers also need ways to contest misuse, attribution, or exclusion from protective tools.
- Explaining a model is technically impossible.
- Recourse does not need a full technical explanation. It needs enough of the basis to contest the decision, and someone with the authority to correct it.
- Appeals are too slow to matter.
- That is a reason to give appeals owners and deadlines, and to ask before a use becomes ordinary, not a reason to have none.