Skip to content

PAN Lab levers

Lever

Require sign-off

The organization requires a named, accountable person to sign off before its AI system goes live. The system also needs that sign-off to stay live. At the strong tier the sign-off scales with the assessed risk, and the riskiest uses wait for a person to adjudicate them.

What it is

A deployment without an owner has no one who can say no. A conformity gate names the person with that authority and makes their sign-off a condition of going live. The framework cited below grades the authority to deploy by measured zones of risk, from routine use with monitoring to suspension.

What it pushes on in the Lab

In the Lab, this lever engages deployment authority, whether someone with real power to change or halt the deployment is involved in it. At its strong tier it also puts a ceiling on the failure regime, because the gate holds back the riskiest operations.

addedDeployment authority engaged

You can also pull this lever at a strong tier, which costs more. At the strong tier, each effect below the Lab's strongest setting pushes harder.

The strong tier also pushes on these:

cappedFailure regime

The Lab applies this lever to the whole network. It acts only on gauges that read the whole deployment, so it has no single part to aim at.

Its pattern in the Practice Library

The Practice Library describes the pattern behind this lever:Conformity assessment gate

The pressures it answers

No pressure page lists this lever among the levers that answer it.

Where you can pull it

The evidence behind its effects

The Lab cites these claims from the evidence registry for this lever's effects.

A frontier risk-management framework in practice ties deployment authority to measured capability-vs-safety zones — green (routine plus monitoring), yellow (controlled with strengthened mitigations), and red (suspend).[3]

shanghaiartificialintelligen2025AcademicSave

Shanghai Artificial Intelligence Laboratory. (2025). Frontier AI Risk Management Framework in Practice: A Risk Analysis Technical Report [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2507.16534

doi.org/10.48550/arXiv.2507.16534

Appears in: PAN framework development

Topics: ai-governance, ai-safety

greenblatt2024AcademicSave

Greenblatt, R., Denison, C., Wright, B., et al. (2024). Alignment Faking in Large Language Models [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2412.14093

doi.org/10.48550/arXiv.2412.14093

Appears in: Evidence reverification (2026)

Topics: ai-alignment, ai-safety

meinke2024AcademicSave

Meinke, A., Schoen, B., Scheurer, J., Balesni, M., Shah, R., & Hobbhahn, M. (2024). Frontier Models are Capable of In-context Scheming [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2412.04984

doi.org/10.48550/arXiv.2412.04984

Appears in: Evidence reverification (2026)

Topics: ai-safety

Pull this lever in the PAN Lab and watch which way it pushes the network.

Open the PAN Lab