Behavry — Compliance & Regulatory Mapping

Informational only. This document is provided for reference purposes. It does not constitute legal or compliance advice. Consult your compliance team, legal counsel, and qualified assessors for formal certifications, audits, or regulatory filings.

---

Overview

Behavry provides technical controls that map to requirements across multiple compliance frameworks, including SOC 2 Type II, ISO/IEC 27001:2022, NIST AI Risk Management Framework (AI RMF), the EU AI Act, GDPR, and HIPAA. The table below summarizes these mappings.

---

Control Mapping

Audit Logging & Monitoring

Control Description	Framework	Clause / Control	Behavry Feature
Log all AI system activity	SOC 2	CC7.2 — System Monitoring	Immutable audit log (TimescaleDB hypertable) captures every tool call, policy decision, DLP finding, and latency measurement
Maintain audit trail integrity	SOC 2	CC7.2 — Log Integrity	SHA-256 hash chain across all audit events; tampering detectable by hash verification
Log and monitor user and system activity	ISO 27001	A.8.15 Logging / A.8.16 Monitoring	Structured JSON logs with agent_id, session_id, tool_name, policy_result, timestamp
Monitor AI-specific behaviors	NIST AI RMF	MG-2.2 — Monitor AI System Behavior	Behavioral Monitor tracks per-agent rolling statistics; anomaly score calculated continuously
Detect behavioral anomalies	SOC 2	CC7.3 — Anomaly Detection	Behavioral Monitor detects statistical deviations; raises ALERT events when thresholds exceeded
Monitor for security events	ISO 27001	A.8.16 Monitoring	EventBus publishes all events to real-time dashboard SSE stream and configurable webhooks

Access Control & Identity

Control Description	Framework	Clause / Control	Behavry Feature
Authenticate all AI agent requests	SOC 2	CC6.1 — Logical Access	Every proxy request requires a valid RS256 JWT issued via OAuth 2.1 client credentials flow
Enforce least-privilege access	SOC 2	CC6.1 — Least Privilege	Role-based permissions (e.g., filesystem:read) enforced at the proxy layer before any tool call reaches the target server
Verify agent identity	NIST AI RMF	MG-2.1 — AI Actor Authentication	Agent identity bound to client_id + client_secret; session tracked by JWT jti claim
Revoke access on termination	SOC 2	CC6.2 — Revocation	Agent suspension and deprovisioning immediately invalidate issued tokens
Rotate credentials	SOC 2	CC6.1	Credential rotation endpoint (POST /api/v1/agents/{id}/rotate) replaces client_secret without downtime
Separate admin and agent identities	ISO 27001	A.8.2 Privileged Access Management	Admin JWT (RS256, separate signing flow) and agent JWT are issued via different endpoints and carry different claims

Policy Enforcement

Control Description	Framework	Clause / Control	Behavry Feature
Enforce technical controls on AI actions	SOC 2	CC6.6 — Boundary Protection	OPA policy engine evaluates every tool call before forwarding; deny results in immediate rejection
Define and enforce acceptable use policies	NIST AI RMF	GV-1.2 — Policies and Procedures	Rego policies define allow/deny/escalate rules per agent type, tool, resource, and action
Maintain policy versioning	SOC 2	CC8.1 — Change Management	Policy version number incremented on each update; history maintained in DB
Test policies before activation	ISO 27001	A.8.8 Management of Technical Vulnerabilities	Policy evaluate endpoint (POST /api/v1/policies/{id}/evaluate) enables pre-activation testing against sample inputs
Prevent unauthorized AI model access	NIST AI RMF	MG-2.2	OpenAI and Anthropic API proxies gate all programmatic model access behind Behavry JWT + OPA policy check

Sensitive Data Controls (DLP)

Control Description	Framework	Clause / Control	Behavry Feature
Protect sensitive data from unauthorized disclosure	SOC 2	CC6.7 — Data Protection in Transit	DLP scanner inspects all tool inputs and outputs; 12 built-in patterns (SSN, credit card, PII, keys, etc.)
Prevent exfiltration of PII and sensitive data	GDPR	Art. 32 — Security of Processing	DLP auto-blocks tool calls rated critical; lower severities raise alerts
Protect PHI / ePHI	HIPAA	§164.312(a)(1) — Access Control	DLP patterns include medical record number and HIPAA-relevant identifiers; critical findings are blocked
Log DLP findings	SOC 2	CC7.2	dlp_findings field in every audit event records all matched patterns with severity
Block outbound data to unauthorized endpoints	SOC 2	CC6.7	OPA policies can deny web write operations to external domains; demonstrated in threat exfiltration demo

Human Oversight of AI

Control Description	Framework	Clause / Control	Behavry Feature
Maintain meaningful human oversight of AI systems	EU AI Act	Art. 14 — Human Oversight	Human-in-the-loop (HITL) escalation queue holds sensitive operations pending human approval
Ensure humans can override AI decisions	NIST AI RMF	GO-1.7 — Human Override	Escalation approve/deny endpoints allow human operators to permit or block held requests in real time
Log human oversight decisions	SOC 2	CC7.2	Escalation resolution (approver identity, timestamp, decision) recorded in audit log
Define escalation timeout policies	EU AI Act	Art. 14	Escalations automatically expire after configurable timeout; timed-out requests are treated as denied
Track escalation audit trail	NIST AI RMF	MG-4.2 — Risk Monitoring	Full escalation lifecycle (created → pending → resolved) emits audit events visible in dashboard

Incident Response

Control Description	Framework	Clause / Control	Behavry Feature
Detect and alert on security incidents	SOC 2	CC7.4 — Incident Response	Alert system raises ALERT events for anomalies, DLP violations, and policy breaches
Notify stakeholders of high-severity events	SOC 2	CC7.4	Webhook delivery module sends signed HTTP POST to Slack, PagerDuty, or custom SIEM endpoints for high/critical alerts
Maintain incident audit trail	ISO 27001	A.5.24 — Information Security Incident Management	All alert events stored with full context (agent_id, session_id, tool_name, details)
Classify alert severity	SOC 2	CC7.4	Alerts are classified as low, medium, high, or critical; webhooks filter by configurable minimum severity
Support threat hunting and forensics	ISO 27001	A.8.15	Audit log queryable by agent, action, policy_result, time range; SSE stream for real-time investigation

Risk Assessment

Control Description	Framework	Clause / Control	Behavry Feature
Assess and score AI agent risk	NIST AI RMF	MG-2.2 — Risk Assessment	Behavry Risk Framework (BRF) risk scorer evaluates 6 dimensions: access, action, decision_scope, oversight, intent_drift, observability
Classify agents by risk tier	NIST AI RMF	GV-6.2 — AI Risk Tiers	Risk tiers (low / medium / high / critical) determine escalation thresholds and policy behavior
Recalculate risk on behavioral change	NIST AI RMF	MG-2.2	Risk score recalculated on demand and after behavioral anomaly alerts
Document AI system capabilities and risks	EU AI Act	Art. 11 — Technical Documentation	Agent registry captures agent_type, owner, assigned roles, and current risk tier

Encryption & Cryptographic Controls

Control Description	Framework	Clause / Control	Behavry Feature
Use strong cryptography for authentication tokens	SOC 2	CC6.1	RS256 (RSA 2048+) JWT signing for all agent and admin tokens
Sign outbound webhook payloads	ISO 27001	A.8.24 — Use of Cryptography	X-Behavry-Signature: sha256=<hmac> on all webhook deliveries; recipients should verify before processing
Never log sensitive credentials	SOC 2	CC6.1	API proxy (X-OpenAI-Key, X-Anthropic-Key) strips and never logs upstream API keys; client_secret stored as bcrypt hash

---

Framework Quick Reference

SOC 2 Type II (Trust Services Criteria)

Behavry primarily addresses the Security trust service category (CC6–CC8) with supporting controls in Availability (CC9) through its reliability design.

CC Category	Description	Coverage
CC6.1	Logical and physical access controls	Identity Service, RBAC, JWT
CC6.2	Account provisioning and revocation	Agent lifecycle APIs
CC6.6	Boundary protection	MCP Proxy, OPA policy enforcement
CC6.7	Data protection in transit	DLP scanner, TLS in production
CC7.2	System monitoring and logging	Audit Logger, hash chaining
CC7.3	Anomaly detection	Behavioral Monitor
CC7.4	Incident response	Alerts, webhooks
CC8.1	Change management	Policy versioning

ISO/IEC 27001:2022

Control	Description	Coverage
A.5.24	Information security incident management	Alert system
A.8.2	Privileged access management	Separate admin/agent identity flows
A.8.15	Logging	Structured audit log
A.8.16	Monitoring activities	SSE dashboard, behavioral monitor
A.8.24	Use of cryptography	RS256 JWT, HMAC webhook signing

NIST AI Risk Management Framework (AI RMF 1.0)

Function	Sub-category	Coverage
GOVERN	GV-1.2 — Policies	Rego policy management
GOVERN	GV-6.2 — Risk Tiers	Behavry Risk Framework risk scoring
MAP	MP-2.3 — Impact Assessment	DLP + alert severity classification
MEASURE	MG-2.1 — Authentication	Agent identity + JWT
MEASURE	MG-2.2 — Monitor Behavior	Behavioral Monitor, anomaly detection
MEASURE	MG-4.2 — Risk Monitoring	Escalation audit trail
MANAGE	GO-1.7 — Human Override	HITL escalation queue

EU AI Act (High-Risk AI Systems)

Article	Requirement	Coverage
Art. 11	Technical documentation	Agent registry with type, owner, risk tier
Art. 14	Human oversight	HITL escalation with approve/deny/timeout
Art. 17	Quality management system	Policy versioning, audit trail, incident response

GDPR

Article	Requirement	Coverage
Art. 5(1)(f)	Integrity and confidentiality	DLP scanner, audit log, access controls
Art. 25	Data protection by design	DLP auto-block on critical findings
Art. 32	Security of processing	Encryption (JWT RS256), DLP, access controls

HIPAA

Section	Requirement	Coverage
§164.312(a)(1)	Access control	Identity Service + RBAC
§164.312(b)	Audit controls	Immutable audit log
§164.312(c)(1)	Integrity	Hash-chained audit events
§164.312(e)(2)	Encryption	TLS in production (infra-level)

---

Evidence Collection Guide

For auditors and compliance teams: the following Behavry artifacts can serve as evidence for the controls listed above.

Evidence Artifact	Location / API	Relevant Controls
Audit log export	GET /api/v1/audit/events (filterable by time range)	CC7.2, A.8.15, Art. 32
Agent registry	GET /api/v1/agents	CC6.1, GV-6.2
Policy definitions	GET /api/v1/policies	CC6.6, GV-1.2
Policy history	GET /api/v1/policies/{id} (version field)	CC8.1
Alert log	GET /api/v1/alerts	CC7.3, CC7.4, A.5.24
Escalation log	GET /api/v1/escalations	Art. 14, GO-1.7
Agent risk scores	GET /api/v1/agents/{id}/risk	MG-2.2, GV-6.2
Behavioral baselines	GET /api/v1/agents/{id}/behavior	MG-2.2

---

Last updated: February 2026