Type-III AI (CRSP-governed agents)
AI that proposes—
institutions that authorize.
PtolemyAI delivers Type-III AI systems built for accountability, auditability, and trust.
This is not “AI productivity.” This is loss prevention + enablement: assistance without authority leakage,
backed by governed release, defeaters, and audit-ready artifacts.
Frontier agent ecosystem: orchestration frameworks (e.g., ADK), tool protocols (e.g., MCP/Skills),
and product “agents” can be operationally powerful—yet remain non-normative.
In CRSP terms, these are typically Type-I executors until outputs are governed by a normative cycle:
Commit → Entitle → Evaluate → Release, and institutionalized via Ask → Evidence → Release (AER)
with defeaters, accountable roles, and audit-ready trace.
Type-I: Generative system without normative awareness. ≠ Type-II: System with guardrails and disclaimers ≠ Type-III: AER + defeaters + policy-trace (institutional), System whose outputs are:
Commitments
Subject to entitlement
Governed by role-based authority
Anchored to institutional policy
Cryptographically attributable
Type-III systems are not “smart chatbots.”
They are:
Norm-governed participants in institutional reason
Sovereign AI for Governments & Regulated Institutions.
Built on CRSP, AER, and Normative Routing Contracts, this approach enables nations and institutions
to adopt frontier AI while retaining legal authority, jurisdictional control, and accountable release.
Global intelligence, locally governed → CRSP context → AER lifecycle → institutionally authorized release
New: Normative Routing Contract (NRC). A partner-friendly intake form that turns “tribal knowledge” into a
repeatable implementation contract. NRC routes each request to the correct Type posture (I/II/III), selects the right
defeaters and evidence rules, and standardizes accountable human roles for governed release.
NRC (intake + posture + accountable roles) → AER (Ask → Evidence → Release) → audit-ready artifacts
Executive thesis: Type-I sells speed. Type-II sells efficiency. Type-III sells institutional survival.
Loss prevention + enablement
Reduce probability and blast-radius of “wrong in public” events: incidents, exams, enforcement scrutiny.
Audit-ready by construction
Artifacts and trace are first-class outputs, not after-the-fact documentation sprints.
Model-agnostic control layer
Decouple governance from LLMs so model churn doesn’t force re-approval from scratch.
Bounded assistance
Drafting without decision authority, advice posture, or autonomy leakage.
Non-advisory posture
Explicit accountable roles
Authority boundary
- Outputs framed as drafts for review and sign-off
- Guardrails tuned to regulated language
- Clear “not a decision-maker” posture
Designed to help humans move faster without making the system “the authority.”
Governed release
Ask → Evidence → Release with built-in caveats, defeaters, and reviewability.
AER gates
Defeater library
Misuse mitigation
- Over-reliance and misuse risk language
- Justification narratives tied to evidence
- Release controls aligned to governance posture
Compliance is embedded in the act of generation—not stitched on later.
Enterprise-native adoption
Land with drafting; expand into a governance standard across AI.
Land → Expand
Procurement-ready
Audit survivability
- Land: documentation, disclosures, checklists, policy drafts
- Expand: governance layer across all AI tools and models
- Displace: copilots that can’t survive exams and audits
Built for how regulated enterprises actually buy, deploy, and defend AI.
Where we land first
Industries where “wrong in public” is existential.
Healthcare
Finance
Government
Defense
Critical Infrastructure
Legal/Compliance
- Patient education, documentation, safety checklists
- Compliance interpretation and audit response drafting
- Transparency and oversight documentation
Start with low-risk drafting; expand governance across your AI portfolio.