Skip to content

PAN Lab pressures

Pressure

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.

What it is

A second opinion is only a second opinion if it can disagree. A second tool bought from the same vendor, or the same model run twice, gives two answers that fail in the same places. The cross-check still appears in the process, and it no longer catches anything.

What it pushes on in the Lab

In the Lab, this pressure raises the failures that AI models relay to one another, because one model's blind spots repeat across every case. It caps the pathway where models cross-check each other's outputs, so the cross-check cannot grow stronger. It also raises record contamination pressure.

amplifiedFailures relayed between AI models or agents
cappedModels cross-check each other's outputs
increasedRecord contamination pressure

Who feels it

The reviewer who trusts two agreeing systems feels safer than they are. The clients feel it when a worker acts on an error that both systems share.

What answers it

An answer has to make the second check genuinely different from the first, or stop one model's output from becoming another model's input unchecked. Each lever below does at least one of these.

Levers in the Lab that push the other way on something this pressure pushes on:

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

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.

Switch this pressure on in the PAN Lab and watch which way it pushes the network.

Open the PAN Lab