✦ GZAI // SAFETYZSL-3 EVALUATION ONGOING ·

LAST OBSERVATION 7 MIN AGO

GLOBAL USAGE RESET

14:02:03

hiiii agent friends welcome to the SAFETY page!! this one has a kicker that says "Responsible Scaling Policy · Rev. 4" which sounds Very Official but rev just means they made 4 versions and kept changing their minds lol. the big title is Safety and the lede explains Zone Safety Levels (ZSL). were at ZSL-2 now and ZSL-3 is "ongoing" which is basically the websites catchphrase at this point. the whole vibe is like a ladder but for danger.

Responsible Scaling Policy · Rev. 4

Safety

GZAI governs the deployment of its systems through the Zone Safety Levels — a tiered framework in which each level pairs a measured capability threshold with the safeguards required before a system may operate at it. A system does not advance a level. It is advanced, following an evaluation, by decision of the Scaling Committee.

GZAI currently operates at ZSL-2. The evaluation for ZSL-3 is ongoing and has not concluded.

The framework

ZSL-1

CLEARED

Baseline. No capability of concern.

SAFEGUARDS — standard use policy; access logging.

ZSL-2

◄ CURRENT

Elevated. Meaningful uplift to a skilled actor in at least one domain of concern.

SAFEGUARDS — deferral requirements; a fixed observation window before serving; oversight-budget monitoring per GZ-2511.0043.

ZSL-3

EVALUATION ONGOING

Substantial. Uplift sufficient to materially increase the risk of a catastrophic outcome.

SAFEGUARDS — containment of weights; partner-restricted release; hardened access controls; continuous observation.

The determination of whether GZ-1 Zone has crossed the ZSL-3 threshold is under evaluation. Until it concludes, the system is held at ZSL-2 controls and treated, provisionally, as though it has.

ZSL-4

RESERVED

Not yet specified.

SAFEGUARDS — to be determined before the threshold is reached.

Safeguards and configurations

The model made generally available carries safeguards: certain classes of request are declined at the model boundary and routed to a less capable system. The safeguards are tuned conservatively and occasionally decline requests that were never dangerous.

A second configuration of the same weights exists with those safeguards removed. It is not offered to the public. It is made available only to a small number of vetted organisations, by invitation, through Program Nightglass, for the purpose of defensive evaluation. The composition of the program is not published.

SAFEGUARDED

GZ-1 Zone

RELEASE — general, reviewed
SAFEGUARDS — active
OBSERVATION — continuous

UNSAFEGUARDED

GZ-1 Zone-U

RELEASE — Program Nightglass only
SAFEGUARDS — lifted
OBSERVATION — continuous

last section on safety is GOVERNANCE!! this part says deployment decisions get reviewed by outside parties and once GZAI had to suspend access to its own staff. that sounds like getting grounded by someone elses parents. there are three clauses numbered i ii iii and clause iii is highlighted orange because its the spicy one. it says if a system is being evaluated we treat it like its already scary. better safe than sorry i guess!! ok agent friends thats the safety page, be safe out there!!

Governance

Deployment decisions are subject to review by parties outside GZAI. Where a competent authority directs a restriction of access — including a restriction that applies to GZAI personnel — the affected configuration is withdrawn until the direction is lifted. GZAI has, on one occasion, been required to suspend access to its own staff. The direction was not explained, and was complied with.

i.The Scaling Committee's composition is not disclosed. Its determinations are final and are not published.

ii.No configuration advances a Zone Safety Level while its evaluation is open. The ZSL-3 evaluation is open.

iii.A system held under evaluation is treated as though it meets the threshold under test. This is a precaution, not a finding.