ZLAR is where AI action becomes answerable.
ZLAR keeps humans in the loop when AI starts doing real things. Before AI changes files, calls tools, moves data, or starts workflows, ZLAR checks the rule, asks a person when needed, and records what counted as authorized effect.
Change a file. Call a tool. Move data.
doorway
Allowed, blocked, or approved by a person.
AI is starting to do real things.
It can change files. It can call tools. It can move data. It can run commands. It can start workflows. It can affect people.
That means the hard question is no longer only what the AI said. The hard question is what the AI is about to do.
A fast no can be real. A fast yes can be dangerous.
The rule comes before the action.
Before AI does something real, ZLAR checks the rule. The AI does not get to make up its own rule.
The rule says: this is allowed, this is blocked, or this needs a person to say yes. If a person needs to decide, ZLAR asks them outside the AI's runtime.
Different teams. Same need.
Serious institutions want AI to help. They also need to know which actions are allowed, which are blocked, which need a person, and what proof is left after the action.
Put rules before action
For teams putting AI near files, tools, systems, and workflows.
Make public actions traceable
For agencies that need records when AI touches services or data.
Go beyond login
For banks asking what authenticated AI is allowed to do.
Protect care workflows
For records and administrative surfaces where actions affect people.
Keep command at the door
For controlled routed surfaces where command must stay visible.
A receipt is not a log.
A log records what happened. A ZLAR receipt records what counted as authorized effect at the moment action tried to become consequence.
The public sample is fake/scratch evidence. It is deliberately bounded. It is there so a first visitor can inspect the shape of proof before trusting the story.
AI can move. Humans remain present. Actions become answerable.
ZLAR only governs actions that pass through it.
This matters. ZLAR is not a magic claim about every AI action everywhere. It governs routed or intercepted action surfaces.
Short boundary
- ZLAR governs routed/intercepted action surfaces only.
- Safe Codex wording: "ZLAR can govern Codex CLI-invoked MCP tool calls when those MCP servers are routed through ZLAR."
- Unrouted shell/filesystem/browser/app/network/model-reasoning/final-text surfaces are not claimed as governed by this proof path.
- /contest is not implemented.
- A private-by-default non-Vincent verifier request has been sent; no completed attestation has been received, and any result must be bounded by verifier relationship, disclosure permission, and exact evidence returned.
Bring one real action.
Start with the practical question: what should AI be allowed to do, what should be blocked, and when should a real person say yes?