What it is
Records outlive the tools that wrote them. A worker today still reads a case note, an assessment, or a flag that a retired system wrote years ago. Then an audit finds that some of those records were wrong. The organization learns that its people have relied on them all along.
What it pushes on in the Lab
In the Lab, this pressure raises two pathways out of the record store. People read the contaminated records and believe them, and the model repeats them as fresh output.
Who feels it
The worker who reads an old record has no reason to doubt it. The client whose case rests on that record carries the error, usually without any way to see where it began.
What answers it
An answer has to make people less likely to believe a contaminated record, or stop the model from repeating one. Each lever below does at least one of these.
Levers in the Lab that push the other way on something this pressure pushes on:
- Mark AI-written recordsPattern:Provenance labeling
- Understand the systemPattern:Understand the system
- Vet connectionsPattern:Connection authorization
The list leaves out levers the Lab has retired, levers it keeps as counter-examples, and any lever no network offers.
Where it starts switched on
Networks in the Lab that start with this pressure switched on:32
- Accelerated Safety Analysis Protocol (ASAP Tool)
- Amazon Flex driver standing and deactivation
- Amazon recruiting engine
- Character.AI crisis-safety stack
- Citi Retail Services Judgmental Review
- Colorado Family Safety and Risk Assessments
- Cost-Proxy Care Stratification
- CrimSAFE criminal-record tenant screening
Every network that starts with it switched on
- EDD Virtual Assistant
- Enova's CashNetUSA and NetCredit loan servicing
- Equifax's Online Model Server
- EviCore by Evernorth prior-authorization screening
- Family-Match (Adoption-Share)
- Google's child-safety detection and account enforcement
- Homebase Risk Assessment Questionnaire
- Intuit's recorded video assessment for promotion
- M-Shwari & Kenya's Digital Credit Market
- Meta employment-ad targeting and delivery optimization
- Oportun Financial Corporation's legal-collections pipeline
- OPTN eGFR Waiting-Time Correction
- Predict-Align-Prevent
- RealPage revenue management
- SafeRent Tenant Screening Score
- Santander Consumer USA's loss forecasting score
- The Digit automated-savings tool, or Oportun Set & Save
- The GIFCT hash-sharing database and member matching system
- The NCMEC CyberTipline reporting and triage system
- TransUnion OFAC Name Screen
- Udbetaling Danmark data-driven control (Denmark)
- United Behavioral Health's Level of Care Guidelines
- US Birth Match
- YouTube's Content ID copyright matching system
The evidence behind its effects
The Lab cites no claim from the evidence registry for this pressure's effects. The Lab's authors set the direction and size of each effect.