Worked Examples
The Predatory-Steward Diagnostic Matrix is not a scorecard. It is a way to structure inquiry: what evidence would suggest predatory drift, what would suggest stewardship, and where does uncertainty remain?
These examples are illustrative rather than definitive. The frontier AI example uses cautious public-evidence categories, not a verdict on any particular company. The civic institution example is a composite meant to show how the matrix applies outside technology.
Example 1: Frontier AI Labs in 2026
Technological Governance
The central question is whether frontier systems are governed before deployment or after harms appear. A steward trajectory would show independent safety evaluation, public reporting, meaningful liability, and credible pacing mechanisms before capability jumps are released into broad use.
Evidence to look for: Evidence to look for includes safety budgets relative to capability spending, third-party audits, incident disclosure, model-risk documentation, and whether governance has authority to delay launches.
Uncertainty: The hardest uncertainty is distinguishing genuine restraint from safety theater, especially when the most relevant evidence is proprietary.
Power Distribution
The matrix asks whether consequential power accumulates among actors insulated from its harms. Frontier AI tends toward concentration because compute, data, talent, and deployment channels are expensive and centralized.
Evidence to look for: Evidence to look for includes ownership concentration, cloud dependencies, API gatekeeping, public-interest representation, and who has recourse when deployments cause harm.
Uncertainty: The open question is how to distribute accountability without making governance so diffuse that no one can act competently.
Knowledge Orientation
A steward trajectory would reward epistemic honesty about capabilities, limits, and risks. A predatory trajectory would turn uncertainty into marketing advantage or suppress inconvenient findings.
Evidence to look for: Evidence to look for includes publication practices, whistleblower protections, model evaluations, red-team transparency, and whether safety claims are independently reproducible.
Uncertainty: Some secrecy may be justified for security. The question is when secrecy protects the public and when it protects the developer.
Time Orientation
Frontier AI compresses time when launch timing is governed by quarterly competition, fundraising cycles, or rival labs. It extends time when decisions account for lock-in, downstream labor effects, dependency, and institutional adaptation.
Evidence to look for: Evidence to look for includes deployment pacing, long-horizon risk planning, sunset clauses, staged access, and explicit assessment of irreversible dependencies.
Uncertainty: The main uncertainty is whether slower deployment reduces risk or simply shifts development to less accountable actors.
Intergenerational Obligation
The strongest steward claim would show that developers preserve future options rather than creating infrastructures future societies cannot meaningfully refuse.
Evidence to look for: Evidence to look for includes reversibility, data governance, environmental costs of compute, institutional dependence, and safeguards against permanent concentration of cognitive infrastructure.
Uncertainty: It remains unclear how to represent future users and non-users in present technical decisions.
Example 2: A Civic Institution or University
Knowledge Orientation
A civic institution or university is steward-oriented when it protects inquiry even when findings are inconvenient to donors, administrators, political coalitions, or prestige.
Evidence to look for: Evidence to look for includes academic freedom protections, publication independence, transparent conflicts of interest, and support for dissenting expertise.
Uncertainty: The difficulty is balancing truth-seeking with legitimate duties of care, privacy, and community trust.
Resilience Investment
A steward institution maintains buffers before crisis: financial reserves, redundant systems, local capacity, mental-health support, and continuity planning.
Evidence to look for: Evidence to look for includes preparedness budgets, crisis drills, staff redundancy, reserve policies, and whether recovery protects the vulnerable first.
Uncertainty: The unresolved question is how much efficiency to sacrifice for resilience, especially under budget pressure.
Cultural Humility
A steward institution learns from traditions and communities outside its dominant culture without extraction, tokenism, or romanticization.
Evidence to look for: Evidence to look for includes consultation quality, shared authorship or benefit, language access, retention of non-dominant knowledge systems, and repair of historical harms.
Uncertainty: The hard question is how to distinguish genuine epistemic humility from symbolic inclusion.
Power Distribution
The matrix asks whether affected people have real standing in decisions that shape their conditions. A steward institution does not confuse listening sessions with shared power.
Evidence to look for: Evidence to look for includes governance seats, budget authority, grievance mechanisms, transparency, and whether participation changes outcomes.
Uncertainty: The tension is preserving competent administration while preventing participation from becoming ornamental.
For the full matrix and its academic framing, see Section 3 of the paper.