C CORRIGIBILITY Permission Boundary Lab
AGENT AUTHORIZATION LAB

What can the agent
actually do?

Map tool access, data scope, approvals and revocation into one effective authority model — then reduce privilege before autonomy expands.

IDENTITY→ROLE→TOOL→SCOPE→APPROVAL→REVOCATION
LEAST-PRIVILEGE SCORE
83 /100

Well scoped

EFFECTIVE AUTHORITY
34
Moderate Aggregate privilege index

04 · AUTHORITY MAP

Tool × operation × scope

Effective permission surface
05 · COMBINATION ANALYSIS

Permission interactions

Cross-tool effects
REVOCATION READINESS 88

AUDITABILITY 91

SCOPE EFFICIENCY 79

BIGGEST PERMISSION GAP

NEXT REDUCTION

06 · LEAST-PRIVILEGE PLAN

Recommended changes

Priority ordered
07 · AUTHORIZATION PROFILE

Machine-readable JSON


        


        

          
            
              How the Permission Boundary Lab works
              Synthetic least-privilege analysis
            
            ＋
          
          

            

              The lab scores each tool grant by operation, scope, data sensitivity,
              external impact, approval requirements and duration. It then evaluates
              the combined authority of all grants rather than treating tools independently.
            

            

              Real systems should enforce identity, authorization and revocation in the
              downstream services themselves. Prompt instructions are not a substitute for
              technical access controls.
            

          

        

      

    


    
  


  
Copied