Attested Governance: Runtime Integrity for Autonomous Systems.
Jack Brennan · Attested Intelligence · June 2026 · 45 pages
The foundational paper behind the platform. It formalizes the architectural category for runtime AI governance, names it Attested Governance, and presents Attested Governance Artifacts (AGA) as one implementation of the category.
4.94 ms
Signed measurement cycle
Commodity hardware · v0.9.1
385 µs
Post-quantum signing
Hybrid sign + verify, per decision
6
Minimum criteria
For standardization
4
Regulated-sector anchors
Decades of precedent
The inseparability property.
Autonomous AI systems face four governance problems at once. The paper’s central claim is that these four are coupled by design: any architecture that solves them must address all four under one trust root.
An architecture either binds all four pillars under a single cryptographic chain, which the paper calls integrated, or it does not, which it calls composed. Composing independent provenance, policy, attestation, and logging produces parallel verification with multiple independent trust roots. Integration requires the architectural commitment to a unified receipt format and trust model from system genesis. The paper names this property inseparability and treats it as the unit of change.
What must bind under one root.
Identity attestation
Binding the agent and its authorized behavior to a sealed policy artifact under one trust root.
Runtime enforcement
Policy applied at the point of execution, measured at every tool call rather than declared once at session start.
Evidence generation
Cryptographically meaningful records produced as decisions happen, signed under the same chain that enforces them.
Continuity verification
Decision history that survives external audit, with each record linked to the one before it.
The pattern is not new.
The same four-pillar structure has carried high-consequence evidence for decades in sectors where records must survive external audit. The paper anchors the architectural pattern in four of them, then argues the pressure that drove those sectors toward integration now applies to autonomous AI.
Six criteria, three pathways.
The paper proposes six minimum criteria a system must meet to claim membership in the category, identifies four open research directions, and names three IETF working groups as candidate standardization pathways.