PAN Lab
Pressures
A pressure is a change in an organization's conditions that its AI governance did not choose. More cases arrive than the staff can check. A vendor updates the model without notice. Workers adopt a tool that nobody approved.
In the PAN Lab, each pressure acts on named parts of a network: the people using the system, the model, the records, and the pathways between them. You switch a pressure on and watch which way it pushes. Each pressure below has a page of its own with what it is, what it pushes on, who feels it, and which levers answer it.
Pressures
Every pressure the Lab can apply
Open a pressure to read what it pushes on, where it starts switched on, and which levers push back.
Workload surges
Workloads rise, and the hours to check the AI system's output do not. Each worker has less time to question what the system hands them, so they accept more of its output as it stands.
Silent vendor update
The vendor changes the model and tells no one. The controls the organization tuned to the old version now govern a different system. On a vendor-hosted deployment, the update can also change what the model sends back to the vendor.
Staff turnover
Experienced workers leave, and the people who replace them have only ever worked alongside the AI system. The new staff have less judgment of their own to set against its output, and they defer to it more.
Contaminated records surface
An audit finds that part of the record system was contaminated long ago. Every person and every model that reads those records has been reading errors as if they were facts.
Autonomy expands
Leadership lets the AI system do more without a person's sign-off. The system writes more directly into the record, and the people around it check less of what it does.
Monitoring goes stale
The dashboards still run, but nobody has to act on what they show. Nobody revises the alert levels set at launch, nobody holds the scheduled reviews, and nobody reads the alerts. Drift builds while the controls stay the same.
Users push back
People lean on the AI system with leading questions and stated positions, and the system bends its answers toward them. An answer that agrees with the person asking is the easiest one to accept.
AI-literacy gap widens
Staff use the AI system more than they were ever trained to judge it. Their checks weaken for lack of skill rather than lack of time, and trust in the tool fills the gap.
Connectors sprawl
A productivity suite the organization approved adds connections between its systems faster than anyone reviews them. The connectors copy unverified content into the record systems and copy records out to the vendor's cloud. Someone authorized every connection once, and nobody has reread the list since.
Conformity spreads
One trusted colleague's way of using the AI system becomes everyone's way. Asking a peer to double-check starts to feel like an accusation. The group's confidence does the work that checking used to do.
Agent links sprawl
Teams connect AI agents to other agents faster than anyone reviews the connections. One agent's output becomes another agent's input without a check, so every agent downstream repeats the first one's error. Each hand-off also copies client details into another agent's context.
Second model is a clone
The organization checks one AI model with a second one, but the second is a copy of the first. Its agreement looks like independent confirmation. A copy shares the original's blind spots, so the same errors pass the check in every case.
Unsanctioned AI use
Under time pressure, staff paste case details into consumer AI tools that nobody vetted. Client information leaves the governed system, and workers copy the tools' answers into notes and records without a label.
Expectations outrun the gains
The time the AI system saves is spent before anyone feels it. Targets rise to match the new pace, so staff work hotter and skim the checks they used to make.
The levers that answer these pressures each have a page in the Practice Library. The Lab Index lists every network the pressures act on.
A pressure is easiest to understand when you watch it act on a network.
Open the PAN Lab