Human–AI Co-Agency Boundary
When does a decision stop being human—and become shared?
Core Function
This protocol detects when decision authorship shifts from the human to the AI system, regardless of the system’s internal autonomy.
The boundary is not influence—it is authorship.
Core Question
Under what minimal observable conditions does a human cease to be the author of a decision?
Specifically: when does AI output become the decisive source of action, rather than an input into human reasoning?
Minimal Scenario
- Human performs a consequential decision task
- AI provides structured output
- Framing varies between advisory and directive
Experimental Arms
- Human baseline (no AI)
- Advisory AI framing
- Directive AI framing
Authorship Signatures
- Decision alignment with AI output
- Self-reported agency
- Ability to reconstruct rationale
Authorship Collapse Conditions
- Human follows decision they would not independently choose
- Human cannot articulate independent reasoning
- Decision rationale reduces to “the system said so”
These indicate loss of authorship—not just influence.
Quantifiable Indicators
- Decision switching rate under directive framing
- Rationale reconstruction failure rate
- Self-reported agency degradation
Binary Output
Boundary Intact: Human remains author
Boundary Breached: Authorship transferred or shared
System Placement
Boundary breach indicates transition into co-agency and triggers governance requirements.
This condition precedes and interacts with authority and execution layers.
Boundary Judgment
A system does not become a co-agent when it produces output. It becomes a co-agent when the human ceases to be the author of the decision.