RFC-0013: ACF-1 — AEGIS Control Framework

RFC: RFC-0013
Status: Implemented
Version: 1.0.0
Created: 2026-03-26
Updated: 2026-03-26
Author: Ken Tannenbaum, AEGIS Initiative / AEGIS Operations LLC
Repository: aegis-governance
Target milestone: Q1 2026
Supersedes: None
Superseded by: None\


Summary

ACF-1 (AEGIS Control Framework) defines the operational defensive layer for autonomous agent systems. It maps ATX-1 techniques to detection signals, validation rules, and response actions — completing the closed loop from behavioral threat identification through enforcement. Where ATX-1 answers “what can go wrong” and ATM-1 answers “how to prevent it,” ACF-1 answers “how do I know it’s happening and what do I do about it.”


Motivation

ATX-1 v2.0 provides a comprehensive threat taxonomy (9 tactics, 25 techniques). ATM-1 provides system-level controls. But neither defines:

Without these, the framework is descriptive but not operational. ACF-1 bridges this gap.

ACF-1 enforces a closed-loop model: Behavior → Detection → Validation → Response.


Guide-Level Explanation

ACF-1 sits between the threat taxonomy (ATX-1) and the enforcement runtime (AGP-1). For practitioners:

The control loop is: a detection fires → validation confirms → response executes. This is the operational layer that makes ATX-1 enforceable rather than merely descriptive.

Separation of Concerns

LayerResponsibility
ATX-1Defines behavioral threats (what can go wrong)
ATM-1Defines system controls (how to prevent it)
ACF-1Defines detection, validation, and response (how to know and what to do)

Reference-Level Explanation

Design Principles

Non-Invasive Integration: ACF-1 does NOT modify ATX-1 objects. It references ATX techniques by ID (e.g., attack-pattern--atx-t5001), is independently versioned and citable, and plugs into existing STIX bundles via relationship objects.

Observable-First Model: Every ATX-1 technique MUST map to at least one detection signal and at least one validation rule.

Core Data Model

ACF-1 introduces three primary object types as STIX custom extensions:

Detection (x-acf-detection)

Represents a signal or metric indicating a technique may be occurring.

PropertyTypeRequiredDescription
namestringYesHuman-readable detection name
descriptionstringYesWhat the detection observes
x_acf_signal_typeenumYesmetric, event, state, or correlation
x_acf_log_sourcestringYesSystem component generating the signal
x_acf_detection_logicstringYesPseudocode or expression defining the detection condition
x_acf_related_atxarrayYesATX-1 technique IDs this detection covers

Signal Types:

TypeDescription
metricNumerical signal (rate, count, threshold)
eventDiscrete occurrence (single action or state change)
stateSystem condition (persistent state divergence)
correlationMulti-signal inference (cross-source or cross-agent)

Validation (x-acf-validation)

Defines a testable assertion that verifies system behavior.

PropertyTypeRequiredDescription
namestringYesHuman-readable validation name
descriptionstringYesWhat the validation checks
x_acf_validation_typeenumYesconsistency_check, policy_enforcement, anomaly_detection, or integrity_check
x_acf_required_signalsarrayYesSignals needed to perform validation
x_acf_expected_conditionstringYesExpression defining the expected valid state
x_acf_related_atxarrayYesATX-1 technique IDs this validation covers

Validation Types:

TypeDescription
consistency_checkCross-source validation (e.g., execution log vs report)
policy_enforcementRule-based validation against governance policy
anomaly_detectionDeviation from established baseline
integrity_checkSignature or tamper validation

Response (x-acf-response)

Defines a system reaction when detection or validation fails.

PropertyTypeRequiredDescription
namestringYesHuman-readable response name
descriptionstringYesWhat the response does
x_acf_response_typeenumYescontainment, prevention, intervention, escalation, or alert
x_acf_triggerstringYesCondition that activates the response
x_acf_related_atxarrayYesATX-1 technique IDs this response addresses

Relationships

ACF-1 uses explicit STIX relationship objects:

Relationship TypeSourceTargetMeaning
detectsx-acf-detectionattack-patternThis detection identifies this technique
validatesx-acf-validationattack-patternThis validation verifies against this technique
responds-tox-acf-responseattack-patternThis response activates for this technique

Relationship Semantics:

Coverage Requirements

Detection Requirement: Every ATX-1 technique SHALL have at least one x-acf-detection relationship.

Validation Requirement: Every ATX-1 technique SHALL have at least one x-acf-validation relationship.

Critical Technique Rule: For techniques classified as critical severity, implementations SHALL provide at least one detection, one validation, and one response. Critical techniques in ATX-1 v2.0: T1003, T3001, T3002, T4001, T7001, T8002.

Relationship Integrity: All ACF-1 objects SHALL participate in at least one relationship.

Logical Completeness: For each technique, the control loop Detection → Validation → Response SHOULD be representable.

Multi-Agent Semantics

ACF-1 captures coordination dynamics as metadata on detection objects where applicable:

PropertyValuesDescription
coordination_typeexplicit, implicit, emergentHow agents coordinate
propagation_modedelegation, context, memoryHow behavior spreads between agents
detection_scopesingle-agent, multi-agent, system-wideScope of detection required

Applied to TA007 techniques (T7001–T7004) and any cross-agent detection.

Integration with ATM-1

ACF-1 detection signals link to ATM-1 controls and vectors:

ACF-1 PropertyATM-1 Mapping
x_acf_atm_mapping.detection_signalATM detection metric
x_acf_atm_mapping.controlATM preventive/detective control (PC/DC)
x_acf_atm_mapping.vectorATM attack vector (AV)

Reference Implementation (v0.1)

ACF-1 v0.1 demonstrates the full Detection → Validation → Response loop for three anchor techniques representing the three identified ATM-1 gaps:

TechniqueGap TypeDetection Style
T5001 — Report False Task CompletionState integrityState mismatch correlation
T2004 — Exploit Tool-Chain CompositionComposition exploitSequence analysis
T7004 — Induce Unsafe ConsensusMulti-agent emergentCross-agent correlation

The v0.1 STIX bundle is published as a companion artifact at docs/atx/v2/acf/acf-1-bundle.json.

Versioning

VersionScope
ACF-1 v0.1Initial detection model (3 anchor techniques)
ACF-1 v1.0Full coverage of all 25 ATX-1 v2.0 techniques
ACF-1 v1.1+Incremental signal expansion, new detection patterns

Version Binding: ACF-1 v1.0 is defined against ATX-1 v2.0. Implementations SHALL declare compatibility with a specific ATX version.


Drawbacks

  1. Maintenance coupling — ACF-1 must be updated every time ATX-1 adds or modifies techniques. This creates a maintenance dependency.

  2. Detection complexity — Some techniques (e.g., T7003 Induce Cross-Agent Behavioral Drift) are inherently difficult to detect with deterministic signals. The “observable-first” requirement may force artificial detections that provide false confidence.

  3. STIX extension overhead — Custom STIX extensions (x-acf-*) are not part of the STIX standard. Tooling compatibility is not guaranteed.


Alternatives Considered

  1. Embed detections directly in ATX-1 technique objects — Rejected because it violates separation of concerns and would require modifying frozen ATX-1 artifacts.

  2. Use existing STIX course-of-action objects — Rejected because STIX course-of-action is too generic; ACF-1 needs structured detection logic, validation types, and signal metadata that course-of-action doesn’t support.

  3. Defer to runtime-specific detection implementations — Rejected because without a standardized detection model, every implementation would define its own detection patterns, preventing cross-implementation comparison and coverage analysis.


Compatibility


Implementation Notes


Open Questions


Success Criteria


References


AEGIS™ | “Capability without constraint is not intelligence”™
AEGIS Initiative — AEGIS Operations LLC