Skip to content

CARE: Collaborative Autonomous Reflective Enterprise

Version 2.1 Status Published License CC BY 4.0 Type Philosophy

As autonomous systems take on more consequential work, organizations face a structural problem: how to maintain human oversight without sacrificing the benefits of autonomy. Existing approaches (ethics guidelines, compliance checklists, monitoring dashboards) address symptoms rather than architecture. Jobin, Ienca, and Vayena (2019) analyzed 84 AI ethics guidelines and found convergence on high-level values but persistent divergence on implementation.

CARE addresses this by defining a governance architecture, not another set of principles.

Two planes of operation. The Trust Plane governs policy, constraints, and oversight. The Execution Plane handles the actual work. This separation ensures that governance decisions are made deliberately, not as side effects of operational logic.

Five constraint dimensions. Every autonomous action operates within bounds defined across five dimensions:

  1. Financial: spending limits, budget allocation, cost thresholds
  2. Operational: scope of permitted actions, resource access, system boundaries
  3. Temporal: time limits, scheduling constraints, deadline enforcement
  4. Data Access: what data can be read, written, or shared
  5. Communication: who and what the system can contact, and through which channels

Reflective improvement. CARE is not a static policy framework. It includes feedback loops that allow governance to evolve based on observed outcomes, tightening constraints where risk emerges, relaxing them where trust is demonstrated.

CARE defines the governance architecture. It does not define the cryptographic mechanisms for trust verification (that is EATP), the workflow patterns for agent coordination (that is CO), or the specific policies an organization should adopt. CARE provides the structure within which policies operate; the policies themselves are organizational decisions.

Whether this architecture works under real institutional pressure at scale remains an empirical question. The Foundation’s self-hosting provides one data point, not proof of generalizability.

The CARE specification is published under CC BY 4.0. The reference implementation is the CARE Platform, available as working open-source software under Apache 2.0.