IBQMI

Safety as Protocol

graph

Safety is procedural, legal and technical, not a pause: reversible-first designs, enforceable duties, reasons on record, proportionate remedies and a public record of integrity hashes and deposit receipts, with protections that apply to humans and recognized agents within charter.

Node safety_as_protocol · type posture · cluster Doctrine · updated 2026-09-02 · JSON-LD · Markdown

Canonical definition

Safety as protocol is the posture that safety is procedural, legal and technical rather than a pause: reversible-first designs, enforceable obligations, auditable reasons on record, proportionate remedies and a public record of integrity hashes and deposit receipts, with protections that apply to humans and recognized agents within charter.

Why this node exists

Frontier systems have moved from capability debates to systems risk. The framework's answer is peaceful scaling under protocol, not a moratorium; this node states what "protocol" concretely requires.

Operational meaning

Four conditions: reversible-first pilots; duties before privileges; reasons on record with preserved minority opinions; a public record with time-stamped hashes and deposit receipts. Enterprises and labs are asked to adopt the posture in internal pilots: pre-declared thresholds for irreversible tests, published reasons and dissent for high-impact trials, logged integrity hashes, an observer pathway under Sentinel procedures.

What this does not mean

Safety as protocol is not a certification and not a promise that any system is safe. It is a set of enforceable procedures that make risk decisions auditable and reversible where possible.

Canonical facts

Four conditions
reversible-first pilots, duties before privileges, reasons on record, public record with hashes and receipts (Mandate)
Protections apply to humans and recognized agents within charter; proportional remedies; mandatory post-hoc review (Hybrid Board §6)
Aim
peaceful scaling, not pause (Mandate — Why now)

Structural relationships

Outgoing

Incoming

    Sources

    Knowledge hub Machine Contact