OSCA
Origin, Semantic, and Consequence Authority
A Multi-Tier Authorization Architecture for Autonomous AI Agents
Why Single-Tier Authorization Fails
Autonomous AI agents propose and execute actions. A single authorization check (who are you?) cannot detect an agent that is authenticated but acting outside its declared scope, or one that declares benign effects but produces harmful ones. OSCA separates authority into three independently verifiable tiers.
Origin
Who authorized the action? Identity, credential, attestation, timestamp, nonce.
Semantic
What does the action mean in context? Intent, scope, constraints, preconditions.
Consequence
What does the action produce in the world? Effects, side effects, reversibility.
No Dimension Guarantees Another
Independence
Each tier is evaluated without trusting the others.
Non-Override
No tier can silently override another. All three must pass.
Atomicity
Execution proceeds only when all three tiers authorize. No partial execution.
Traceability
Every authorized action carries an auditable decision trace.
Phase I · Specification Complete
View the Implementation
Live dashboard of tests and scenarios, detailed results, and full documentation.
Open Dashboard →