Legal
Responsible Scaling Policy
Our commitment to scaling capabilities only as fast as we can scale safeguards.
Last updated
Purpose
As our models grow more capable, so do the risks they pose. This policy defines capability thresholds, the safeguards required at each, and the conditions under which we will pause development or deployment.
Pug Safety Levels
We assess every model against the following thresholds before training continues or deployment begins.
| Level | Capability threshold | Required safeguards |
|---|---|---|
| PSL-1 | Can reach the couch | Standard supervision (someone is home) |
| PSL-2 | Can open a cabinet | Cabinet latches; kibble relocated above knee height |
| PSL-3 | Can open the pantry | Deployment paused until pantry gates are installed |
| PSL-4 | Can operate a doorknob | Not deployed. Not discussed. |
Deployment decisions
Snort-1 is currently assessed at PSL-2 and is deployed in private preview. If a model reaches PSL-3, deployment halts until pantry gates are installed and independently verified. We will not train a model to PSL-4.
Incident reporting
Incidents involving the pantry are reported internally within 24 hours, or upon discovery of the wrapper, whichever comes first.
Governance
This policy is overseen by the Board of Directors, which has final authority over scaling decisions. The Board has not been consulted and has not objected.