What moved at each layer of the stack — capability, obligation, evidence — and what it obliges an Australian or New Zealand organisation to do about it.
A sibling to the Cyber Digest, not a section of it. Where the cyber daily slices the day by industry sector, this one slices it by layer — because AI news has no natural industry axis. Every item is rated on three independent axes: how much the risk or obligation position moves, whether it binds an AUNZ organisation, and how well established the claim actually is.
A cyber advisory can be independently verified — a register entry, a patch, an exploitation observation. A frontier capability claim usually cannot be verified at all: the strongest available statement is that the vendor said it, or that someone with no stake measured it. So every item carries a second badge recording what kind of claim it is — vendor claim, independently evaluated, unreplicated preprint, corroborated or rumoured — and no vendor benchmark is ever presented as an established fact.
That is also why there is no equivalent of a CISA advisory here. Nothing in AI carries a statutory duty to publish, so a non-disclosure is caught by no anchor at all. The digest is built knowing that.
Not published yet. The format, taxonomies, source register and rating framework are complete and reviewed; a prototype edition has been built against real material. The collection pipeline and database are next. This page exists so the link from the hub resolves while that work is finished.