Physix Frontier · News Briefing Card (Hacker News · Nov 21, 2026)

Study dissects 25 authorization documents to guide AI agent delegation design

KEY FACTS

  • Author Dazza Greenwood released version 0.4 of the study on November 19, 2026, breaking down 25 authorization documents one by one.
  • Each document is analyzed uniformly across five dimensions: title, background, authorization clauses, structure, and insights.
  • The study distills 12 categories of authorization elements, including principal identity, agent credentials, scope, revocation, and records.
  • The author notes that questions around scoping AI agent authorization, revocation, and fund transfers already have ready-made answers.
  • The study stresses that the key design work lies mostly in registration, notification, and logging mechanisms outside the authorization clauses themselves.
  • California's CCPA rules for authorized agents require businesses to verify directly with consumers on high-risk requests.

KEY DATA

25Documents dissected
0.4Study version
12Authorization element categories

PHYSIX OBSERVATION

This study treats traditional authorization documents as ready-made teaching material for AI agents, a pragmatic approach. What truly deserves attention is risk-tiered verification: low-risk exit-type operations must not face extra hurdles, while only high-risk operations need re-confirmation with the principal. Teams building agent authorization should prioritize borrowing registration, logging, and revocation mechanisms rather than fixating solely on authorization wording.

Source: Hacker News report

Study dissects 25 authorization documents to guide AI agent delegation design | Physix Frontier Briefing