nhi.sgit.ai / documents / ambient-authority

Plugins Are Capability Grants, Not Modules: Ambient Authority Is Why A Prompt Injection Works

TypeArch brief Versionv0.33.56 Date6 August 2026 AuthorDinis Cruz (project lead) and collaborators LicenceCC BY 4.0 Sourceraw markdown · view on GitHub

Summary

Ambient authority — components exercising authority they simply have, rather than authority explicitly granted — is the documented root cause of prompt injection: the attacker asks the deputy to use authority it already holds. A capability is an unforgeable reference to a resource together with permissions over it, granted explicitly; organised by authority rather than by code, plugins become a security boundary. Capabilities should attenuate but never expand, so delegation narrows by construction — and you can instrument every grant before you enforce, turning migration into measurement.

Key concepts

Key ideas

On this site

Curated under 'what is an agent actually allowed to do?' in the collection; the root-cause frame behind the thesis's evidence.

Infographic

Slot reserved. The matching LinkedIn infographic has not yet been linked — the request list is on the infographics page (N1 on comms). Once identified it will appear here, pointing back to this document.

Read the document

📄 Original document · v0.33.56 · 6 August 2026 · rendered from the raw markdown (the source of truth)