]> Violet Shores Pty Ltd

rob@violetshores.com

Security agent authorization assurance audit policy control law Governance of autonomous agents today is largely expressed as boundary enforcement: an action is permitted or blocked at the point it is attempted, per a policy evaluated at that boundary. As agents span heterogeneous action types — authenticating a human, executing a delegated task, selecting a computational resource — a single, uniform way to express "how much assurance this action requires, before it proceeds" is missing. This document describes an interface for pre-action, risk-graded assurance: a policy stage that, before an agent action proceeds, derives an assurance requirement from a risk signal and expresses that requirement in a domain-appropriate form, recording the decision in an audit record and optionally binding it to a verified human root. It defines the interface and the audit-record fields, not any particular risk-scoring method or control law. This document also describes the autonomy-asymmetry control law: a feedback loop coupling assurance requirements to VERIFY-phase pass rates, with fast-down (immediate elevation on failure) and slow-up (hysteresis- governed relaxation on sustained success) asymmetry. The iteration governor is described as the per-packet instance of this control law, and the phase-seal chain as its sensor. This document is offered as input to the proposed AUDIT working group's work on authorization state over time and action provenance.

Introduction Emerging agent-governance systems evaluate policy at action boundaries and return, in the common case, a binary decision. This is adequate for uniform allow/deny gating but does not capture a property that liability-bearing and safety-sensitive deployments increasingly require: assurance should scale with risk, be evaluated before the action proceeds, be expressible uniformly across different kinds of action, and leave a provable record of why a given assurance level was required and whether it was met. Consider three actions an agent may take: (a) re-authenticate the human on whose behalf it acts; (b) execute a delegated task with external effect; (c) select which computational resource (for instance, which model) will handle a request. These belong to different domains, yet each admits a natural notion of "assurance": authentication depth for (a), oversight degree for (b), and selection rigor for (c). Today each is governed by a separate mechanism, if at all. This document describes a single pre-action stage that unifies them at the interface level. It does not prescribe how risk is computed, nor the control law that maps risk to requirement; those are implementation and product matters, and in some deployments proprietary. What it standardizes is the shape of the decision and its record, so that audit consumers and interoperating systems can reason about assurance uniformly. This version (-01) additionally describes the autonomy-asymmetry control law (Section 4) as a reference design for the mapping from risk signal to assurance requirement. The control law is informative; an implementation is free to use any mapping. It is described here because it captures a property — asymmetric response to failure versus success — that recurs across safety-sensitive systems and that the phase-seal chain makes observable.

Relationship to the policy decision/enforcement model The separation of a decision function from an enforcement function is a well-known model in this community: the Common Open Policy Service defined a policy decision point (PDP) and a policy enforcement point (PEP), with the PEP consulting the PDP before acting. The stage described here follows that established division of responsibility: an assurance decision function computes the required assurance for a request, and an enforcement point admits the action only once the requirement is met. What this document adds to that model, for the agent setting, is (a) a requirement that is risk-graded rather than a single admit/deny outcome, (b) evaluation that is uniform across heterogeneous action-domains, and (c) an audit record of the decision suitable for the provenance and authorization-state work of the proposed AUDIT working group. The terminology of is used here for continuity; this document does not reuse the COPS wire protocol.

Relationship to the proposed AUDIT work The proposed AUDIT working group contemplates modeling authorization state over time and action provenance. A pre-action assurance decision is exactly a unit of authorization state: it is computed, applied, and — per this document — recorded, before the action. The audit-record fields in Section 6 are offered for that model.

Relationship to the Verified Human Root Where a deployment also uses a verified human root (), the assurance decision MAY be bound to the root attestation, so that the record shows not only what assurance was required and met, but under which accountable human the action was authorized. The two documents are independent; either may be used without the other.

Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

The Pre-Action Assurance Stage This section describes the decision whose record is the subject of this document. It is descriptive context for the audit-record fields of Section 6, not a mandated enforcement architecture: a deployment records these fields wherever it makes such a decision, however its own enforcement is structured. An implementation that records these fields evaluates, for an incoming agent action request and before the action proceeds: a risk signal for the request (the method of computing it is out of scope); an assurance requirement derived from that signal, expressed in a form appropriate to the request's action-domain (Section 5); whether the requirement is satisfied; and a decision that gates the action at an enforcement point (a PEP in the sense of ): the action proceeds only if the requirement is met, and cannot bypass the stage. The decision, its inputs, and its outcome are recorded (Section 6). The requirement is "risk-graded": it may take one of several ordered levels rather than a single permit/deny outcome. The number of levels, and the mapping from risk signal to level, are out of scope for this document; a deployment MAY use as few as two levels, but an interoperating consumer MUST NOT assume only two levels exist.

Autonomy-Asymmetry Control Law This section describes the autonomy-asymmetry control law: a reference design for the mapping from risk signal to assurance requirement. It is informative. The control law captures the property that loss of autonomy SHOULD be immediate and recovery of autonomy SHOULD be gradual — an asymmetry that reflects the greater cost of a missed safety event versus the modest cost of temporarily elevated assurance.

Fast-Down / Slow-Up Asymmetry The control law applies asymmetric dynamics to assurance requirement changes:

Fast-down (immediate elevation):

When a VERIFY-phase check fails, the assurance requirement for the affected action class MUST step up by one level immediately, without delay or hysteresis. A single failure is sufficient to trigger elevation. Implementations MUST NOT defer or smooth elevation responses.

Slow-up (hysteresis-governed relaxation):

When the VERIFY-phase pass rate rises above a threshold, the assurance requirement MAY step down by one level only after the pass rate has been sustained above that threshold for a configurable number of consecutive evaluation cycles (the hysteresis window). A deployment MUST configure a non-zero hysteresis window; a window of one (immediate relaxation on a single pass) degrades the asymmetry to a symmetric step function and SHOULD NOT be used in safety-sensitive deployments.

The combination — fast-down, slow-up — produces a system in which autonomy is earned slowly through consistent verified performance and lost quickly on any failure. This asymmetry is the defining property of the control law.

VERIFY-Pass-Rate Coupling The control signal for the control law is the VERIFY-phase pass rate, defined as the fraction of recent VERIFY-phase evaluations that returned a passing outcome. The look-back window and minimum sample size are deployment choices; this document does not specify them. The coupling is: pass rate below lower threshold: trigger fast-down (step up requirement); pass rate above upper threshold, sustained for the hysteresis window: enable slow-up (step down requirement); pass rate between thresholds: no change (the hysteresis band). The lower and upper thresholds define the hysteresis band. They MUST be distinct (lower_threshold < upper_threshold); a zero-width band eliminates hysteresis and SHOULD NOT be used. Implementations MAY use a rolling window over recent VERIFY outcomes, a cumulative exponential average, or another estimator. The choice of estimator is out of scope; the coupling to VERIFY outcomes is normative for implementations that claim to implement the autonomy-asymmetry control law.

Iteration Governor (Per-Packet Instance) The iteration governor is the per-packet instance of the control law. It operates at the granularity of a single action packet — a discrete unit of agent work — rather than across a fleet or over a long time horizon. When a packet enters a VERIFY phase and the phase fails, the iteration governor applies fast-down at the packet level: the packet re-enters the RESEARCH phase (the governed re-entry loop), the assurance requirement for that packet's next EXECUTE phase is elevated, and each re-entry is recorded as a seal with a rising iteration counter. When a packet completes N consecutive VERIFY passes (the hysteresis window), the iteration governor permits the next packet in the sequence to begin at a lower assurance requirement (slow-up). The iteration governor thereby instantiates the fleet-level control law at the packet level: each packet's seal chain is the per-packet sensor and the per-packet record of autonomy-level transitions.

Seal Chain as Sensor The phase-seal chain (a SCITT-shaped append-only chain of COSE-shaped signed phase transitions; see Appendix B for an implementation description) is the sensor that feeds the iteration governor and, through it, the control law. The controller reads from the seal chain: the phase of each seal (to identify VERIFY-phase outcomes); the iteration counter (to detect re-entries and their depth); the prev_hash link (to establish chain integrity before reading outcomes); and the spec_contract_hash (to confirm the VERIFY phase ran against the current specification contract, not a stale version). A controller MUST verify chain integrity (prev_hash traversal, hash recomputation) before trusting any seal's content as a sensor reading. A chain whose integrity check fails MUST be treated as if all VERIFY outcomes in that chain were failures (fail closed). This coupling — seal chain as sensor, iteration governor as actuator, fast-down/slow-up as the control law — constitutes the autonomy-asymmetry feedback loop. The loop's state is fully observable from the chain: no out-of-band state is required.

Domain-Appropriate Assurance Forms The assurance requirement is one primitive expressed differently per domain. This document defines three domains and leaves the set extensible.

Human-authentication domain:

The requirement is an authentication depth, ranging from a light single-factor re-authentication for low-risk actions to a full multi-modal re-authentication for high-risk or non-repudiable actions.

Agent-action-execution domain:

The requirement is an oversight degree, ranging from autonomous execution, through draft-then-human-confirmation, to block-pending-human-authorization.

Computational-resource-selection domain:

The requirement is a selection rigor, ranging from single-resource selection to multi-resource cross-validation or consensus of a specified depth.

An implementation MAY define additional domains. A domain definition MUST specify the ordered set of assurance levels for that domain and how a level is determined to be satisfied.

Audit Record Fields For each evaluation, an implementation SHOULD record, in the deployment's audit-record model:

action_ref:

A reference to the action request being gated.

risk_signal:

The risk value that drove the decision, and a reference to the method or metric identity that produced it (the metric's internals are out of scope).

domain:

The action-domain of the request.

assurance_required:

The derived requirement, in the domain-appropriate form of Section 5, including the level within that domain's ordered set.

assurance_outcome:

Whether the requirement was satisfied, and by what evidence reference.

pre_action:

A boolean or structural indicator that the evaluation preceded and gated the action (as opposed to a post-hoc log).

verified_human_root:

OPTIONAL. A reference per binding the decision to an accountable human.

control_law_state:

OPTIONAL. Where the autonomy-asymmetry control law (Section 4) is in use: the assurance level at which the control law stood when this decision was made, the most recent VERIFY-phase outcome that influenced it, and the iteration counter of the current packet. This field makes the control law's state observable in the audit record without requiring out-of-band state.

These fields let an audit consumer answer, for any recorded action: what was the assessed risk, what assurance did policy therefore require, was it met, and was the check made before the action — uniformly across domains. Where the control_law_state field is present, a consumer can additionally answer: at what autonomy level was this action authorized, and what sequence of VERIFY outcomes drove the system to that level.

Interoperability Considerations Two systems interoperate at this interface when they agree on: the set of action-domains in use; for each domain, the ordered set of assurance levels and their satisfaction criteria; and the audit-record field names and encodings. This document fixes the field semantics and the three baseline domains; the concrete level sets and encodings are expected to be profiled by the consuming ecosystem or working group. Systems that evaluate policy only at action boundaries can expose a degenerate two-level grading through this interface without adopting continuous grading internally; consumers therefore MUST treat two-level behavior as a valid special case and MUST NOT infer richer semantics than a producer advertises. Where the autonomy-asymmetry control law (Section 4) is used, the thresholds (lower_threshold, upper_threshold) and hysteresis window MUST be agreed between producer and consumer when they are used to derive the assurance requirement exposed in the audit record. A consumer that does not understand the control_law_state field MUST ignore it rather than fail; the field is OPTIONAL.

Security Considerations

Non-bypass:

The value of a pre-action stage is that it cannot be skipped. An implementation MUST ensure the action does not proceed unless the assurance requirement is satisfied; a stage that can be bypassed provides no assurance regardless of how it grades risk.

Risk-signal integrity:

The assurance requirement is only as trustworthy as the risk signal. Manipulation of the signal manipulates the requirement. Deployments SHOULD protect the integrity and provenance of the risk input; this document's audit fields record which metric was used so that later review can detect anomalous inputs.

Record integrity:

Assurance decisions are security-relevant events. The audit records SHOULD be carried in a tamper-evident record model so that the "required vs met" history cannot be silently rewritten.

Binding to human root:

Where the verified_human_root field is used, the security considerations of apply to that binding, including its replay and revocation properties.

Downgrade:

An attacker who can lower the derived requirement (by influencing the risk signal or the domain classification) weakens assurance without triggering a deny. Deployments SHOULD monitor for anomalous downward drift in required assurance relative to comparable historical actions.

Control-law manipulation:

The autonomy-asymmetry control law (Section 4) relies on the integrity of the VERIFY-phase outcomes read from the seal chain. An attacker who can inject false passing outcomes into the seal chain can accelerate slow-up and obtain a lower assurance requirement without earning it. Implementations MUST verify chain integrity (Section 4.4) before treating any seal as a sensor reading. A chain with a failed integrity check MUST be treated as if all VERIFY outcomes in it were failures.

Hysteresis parameter choice:

A hysteresis window that is too short weakens the slow-up property; a window that is too long makes recovery from a false VERIFY failure disproportionately slow. Deployments SHOULD tune the hysteresis window to the consequence severity of the action class being governed.

IANA Considerations This document has no IANA actions. A future version may register the audit-record field names of Section 6, subject to working-group adoption.

&RFC2119; &RFC8174; &RFC2748; &RFC9635; Violet Shores Pty Ltd

Design Rationale The interface deliberately stops at the decision boundary. Whether risk is computed by a single model, an ensemble, a calibrated cross-model metric, or a human-tuned heuristic is a deployment choice; whether the mapping from risk to requirement is a step function, a continuous curve, or a control loop with feedback is likewise out of scope for the normative parts of this document. The autonomy-asymmetry control law (Section 4) is described here as a reference design precisely because it captures a pattern — asymmetric response to failure — that is well-motivated in safety-sensitive agent systems. The fast-down property reflects that the cost of a missed safety event is typically much higher than the cost of temporarily elevated assurance. The slow-up property with hysteresis reflects that trust, once lost, should be re-established by demonstrated performance rather than by the mere passage of time or a single success. The choice of the phase-seal chain as the sensor (Section 4.4) reflects a design constraint: the control law's state must be fully observable from the chain record alone, without requiring a separate control-state store. This makes the loop's history auditable by any third party with access to the chain, which is consistent with the tamper-evident record requirement of Section 8.

Changes from -00 Added Section 4 (Autonomy-Asymmetry Control Law): the fast-down/slow-up control law, VERIFY-pass-rate coupling, the iteration governor as the per-packet instance, and the phase-seal chain as sensor. No change to the assurance interface or the audit-record fields.

Acknowledgments This document responds to the direction of the proposed AUDIT work on authorization state and provenance, and is designed to compose with .