Capability crosswalk: Falcon Guardian AIDR detection and enforcement + SWT3 independent witness attestation for AI agent governance, Continuous Identity, and shadow AI
For: SOC teams deploying Falcon Guardian, CISOs evaluating independent compliance evidence for AIDR, security architects mapping Falcon telemetry to EU AI Act and NIST AI RMF, auditors assessing AI agent governance controls
Time: 15 minutes to read; 2 hours to integrate
Context: CrowdStrike announced Falcon Guardian at Fal.Con 2026 (Sep 1-3, Las Vegas). This guide maps its 7 core capabilities to 29 SWT3 independent witness procedures.
CrowdStrike Falcon Guardian is an AI Detection and Response (AIDR) solution announced at Fal.Con 2026. It delivers runtime visibility and enforcement for AI agents across enterprise endpoints, SaaS platforms, and cloud environments.
Falcon Guardian operates as the detection and enforcement layer:
Falcon Guardian integrates with Falcon IQ (50+ pre-built agentic AI agents on NVIDIA Nemotron models), Charlotte AI AgentWorks, and the CrowdStrike Falcon MCP Server for AI-native security operations.
Falcon Guardian handles detection and enforcement: discovering agents, monitoring execution graphs, blocking unauthorized behavior, and revoking access in real time. This is essential. But regulated enterprises face a second requirement: independent attestation that governance controls were active, effective, and unmodified during operation.
SWT3 fills this gap by operating as an out-of-band witness layer:
The two layers are complementary: Falcon Guardian detects and enforces at runtime. SWT3 attests that detection and enforcement happened, independently.
| Falcon Guardian Capability | Guardian Detects / Enforces | SWT3 Independently Attests | Procedures | Coverage |
|---|---|---|---|---|
| AI Agent Discovery | Discovers known and shadow agents, deployment status, security posture | Agent identity binding, scope attestation, lifecycle event witnessing | AI-ID.1, NHI-SCOPE.1, NHI-CYCLE.1 | Full |
| Runtime Execution Graph | Causal chain: prompt to identity to tool call to system action | Independent tool call witnessing, permission attestation, chain handoff, trust degradation tracking | AI-TOOL.1, AI-TOOL.2, AI-CHAIN.1, AI-CHAIN.2 | Full |
| Agent Access Controls | Which agents can run, block unauthorized, enforce policy | Resource access witnessing, delegation chain attestation, revocation records | AI-ACC.1, NHI-AGENT.1, NHI-REVOKE.1 | Full |
| Runtime Detection and Response | Detects attacks, malicious behavior, blast radius determination | Threat detection attestation, drift detection, consequence-mapped thresholds, emergency override lifecycle | AI-SEC.1, AI-SEC.2, AI-DRIFT.1, AI-DRIFT.2, AI-EMRG.1 | Full |
| Continuous Identity (SGNL) | SPIFFE-based cryptographic identity, real-time context-aware authorization | Hardware attestation, TPM binding, trust verification, credential presentation | AI-HW.1, AI-HW.3, AI-TRUST.1, AI-TRUST.2 | Full |
| Shadow AI Governance | Discovers AI tools, LLM runtimes, MCP servers, IDE extensions across SaaS, browser, cloud | NHI scope, lifecycle, privilege change, rotation, delegation, and revocation witnessing | NHI-SCOPE.1, NHI-CYCLE.1, NHI-PRIV.1, NHI-ROTATE.1, NHI-AGENT.1, NHI-REVOKE.1 | Full |
| MCP Server Discovery | Discovers MCP servers running on managed endpoints | AI bill of materials, supply chain risk attestation | AI-SBOM.1, AI-SUPPLY.1 | Full |
7 of 7 capabilities have full independent attestation coverage across 29 SWT3 procedures in 8 namespaces.
Falcon Guardian detects: Known and shadow AI agents across Windows and macOS endpoints. Provides a live inventory of every running and dormant agent, who deployed it, and its security status.
SWT3 attests: Every discovered agent can be independently witnessed with a cryptographic identity anchor (AI-ID.1). The agent's scope (permissions, TTL) is attested via NHI-SCOPE.1, and lifecycle events (issued, activated, suspended, expired, revoked) are witnessed via NHI-CYCLE.1. An auditor can verify which agents were active at any point in time, independently of the Falcon console.
Regulatory: EU AI Act Art. 12(1), NIST AI RMF MAP 1.1, OWASP Agentic MCP-01/MCP-07
Falcon Guardian detects: The full causal chain from user prompt, through identity and tool call, to every downstream system action. Reconstructs the agent execution graph.
SWT3 attests: Every tool call is independently witnessed (AI-TOOL.1) with a SHA-256 hash of the tool name and serialized arguments, plus the outcome (success or failure). Tool permission attestation (AI-TOOL.2) validates that runtime tools match the agent's charter, detecting permission drift. For multi-agent chains, AI-CHAIN.1 witnesses inter-agent handoffs with chain depth and acceptance status, while AI-CHAIN.2 tracks trust level changes across handoffs.
Regulatory: EU AI Act Art. 14(4), NIST AI RMF GOVERN 1.7, OWASP Agentic MCP-02/MCP-05
# Falcon Guardian traces the execution graph.
# SWT3 independently witnesses every tool call in that graph.
from swt3_ai import Witness
witness = Witness(
endpoint="https://sovereign.tenova.io",
api_key="axm_live_...",
tenant_id="YOUR_TENANT",
agent_id="falcon-monitored-agent",
signing_key="your-signing-key",
)
# wrap_tool witnesses each invocation with AI-TOOL.1
@witness.wrap_tool
def search_database(query: str, scope: str = "read"):
return db.search(query, scope=scope)
# Every call is now independently attested
results = search_database("quarterly earnings", scope="read-only")
Falcon Guardian enforces: Defines which AI agents can run on managed endpoints, blocks unauthorized agents, and converts governance policy into runtime controls.
SWT3 attests: Every access decision is witnessed (AI-ACC.1) with a SHA-256 hash of the resource identifier, the access scope, and the grant/deny outcome. For delegation chains where agents spawn sub-agents, NHI-AGENT.1 witnesses the parent-child credential relationship and delegation depth. When credentials are revoked, NHI-REVOKE.1 records the reason code and whether the revocation cascaded to child agents.
Regulatory: EU AI Act Art. 9(4)(c), NIST AI RMF MANAGE 2.4, NIST 800-53 AC-4, OWASP Agentic MCP-02/MCP-07
Falcon Guardian detects: Attacks on agents (prompt injection, data exfiltration), malicious agent behavior, and determines blast radius in real time. Falcon IQ automates assessment and remediation with 50+ pre-built agents.
SWT3 attests: Adversarial threat detection events are witnessed (AI-SEC.1) with threat category count, detections, and blocked flag. Input validation outcomes are attested (AI-SEC.2). Statistical drift in agent behavior is detected and witnessed (AI-DRIFT.1) with drift type classification. Consequence-mapped drift thresholds (AI-DRIFT.2) trigger graduated responses from notification through circuit breaker to shutdown, each independently attested. Emergency overrides (AI-EMRG.1) are witnessed with lifecycle chain anchors linking the override event to the triggering condition and resolution.
Regulatory: EU AI Act Art. 15(4), NIST AI RMF MANAGE 2.3, OWASP Agentic MCP-03
Falcon Guardian enforces: Every agent is assigned a cryptographically verifiable identity based on the SPIFFE standard (built on the $740M SGNL acquisition). Access is continuously authorized in real time based on agent ownership, caller identity, and device risk posture. Zero standing privilege: access is granted the moment it is needed and revoked the moment it is not.
SWT3 attests: A second, independent identity layer complements SPIFFE. Hardware runtime attestation (AI-HW.1) witnesses the execution environment (GPU topology, accelerator memory, silicon vendor). TPM platform attestation (AI-HW.3) binds agent identity to hardware PCR registers with quote verification. Trust verification (AI-TRUST.1) evaluates trust levels from denied through sovereign. Trust credential presentation (AI-TRUST.2) enables cryptographically signed credential exchange for inter-agent authorization.
The two identity layers are complementary: SGNL/SPIFFE proves identity to the CrowdStrike platform. SWT3 proves identity to the auditor.
Regulatory: EU AI Act Art. 9/11/15, NIST AI RMF GOVERN 1.2/1.5, NIST 800-53 SI-7, OWASP Agentic MCP-06/MCP-07/MCP-09
# Dual identity: CrowdStrike SPIFFE platform identity + SWT3 auditor identity
witness = Witness(
endpoint="https://sovereign.tenova.io",
api_key="axm_live_...",
tenant_id="YOUR_TENANT",
agent_id="falcon-protected-agent", # Independent auditor identity
signing_key="your-hmac-key", # HMAC-SHA256 payload signing
signing_algorithm="hmac-sha256", # Or "ml-dsa-65" for post-quantum
)
# Witness hardware environment for the agent runtime
witness.witness_hardware(
gpu_count=4,
accelerator_memory_gb=80,
accelerator_type="nvidia",
)
Falcon Guardian discovers: Unapproved AI tools, local model runtimes, SDKs, agent frameworks, MCP servers, and IDE extensions across endpoints, SaaS platforms (Microsoft Copilot, Salesforce Agentforce, ChatGPT Enterprise, OpenAI Enterprise GPT), browsers, and cloud environments. Normalizes agent attributes across vendors for cross-platform governance.
SWT3 attests: Every non-human identity discovered by Falcon Guardian can be independently witnessed through 6 NHI procedures. NHI-SCOPE.1 attests the permissions and TTL of each credential. NHI-CYCLE.1 witnesses lifecycle state transitions (issued, activated, suspended, expired, revoked). NHI-PRIV.1 records before-and-after scope changes for privilege escalation detection. NHI-ROTATE.1 cryptographically links old and new credential identifiers during rotation. NHI-AGENT.1 maps parent-child delegation trees. NHI-REVOKE.1 records revocation reason codes and cascade flags.
Regulatory: EU AI Act Art. 9, NIST AI RMF GOVERN 1.3/1.5, NIST 800-53 IA-4/IA-5, OWASP Agentic A01
Falcon Guardian discovers: MCP servers running on managed endpoints as part of its shadow AI discovery sweep. Identifies MCP server configurations, tool registrations, and API connections.
SWT3 attests: The AI bill of materials (AI-SBOM.1) independently documents MCP server components, dependencies, model integrations, and infrastructure. Supply chain risk attestation (AI-SUPPLY.1) records the number of assessed suppliers, compliance status, and risk level. Combined, these procedures create an auditable inventory of every MCP server in the environment, independently of the Falcon console.
Note: The SWT3 MCP Witness Middleware provides zero-code AI-TOOL.1 witnessing for any MCP transport, complementing Falcon Guardian's discovery with independent attestation of every tool call routed through MCP servers.
Regulatory: EU AI Act Art. 11, NIST AI RMF MAP 5.2, G7/CISA AI SBOM Guidance
Falcon Guardian and SWT3 operate on separate planes. Guardian handles detection and enforcement on the security plane. SWT3 handles attestation on the compliance plane. Neither depends on the other, but together they provide both runtime protection and independent evidence.
# pip install swt3-ai
from swt3_ai import Witness
# Initialize the independent witness layer alongside Falcon Guardian
witness = Witness(
endpoint="https://sovereign.tenova.io",
api_key="axm_live_...",
tenant_id="YOUR_TENANT",
agent_id="falcon-monitored-agent",
signing_key="your-hmac-key",
)
# Wrap any tool -- witnesses every invocation with AI-TOOL.1
@witness.wrap_tool
def query_database(sql: str):
return db.execute(sql)
# Witness agent access decisions alongside Falcon Guardian enforcement
witness.wrap_access(
resource="customer-database",
scope="read",
granted=True,
)
# Run the demo to see witness anchors in your terminal
# python -m swt3_ai.demo
// npm install @tenova/swt3-ai
import { Witness } from "@tenova/swt3-ai";
const witness = new Witness({
endpoint: "https://sovereign.tenova.io",
apiKey: "axm_live_...",
tenantId: "YOUR_TENANT",
agentId: "falcon-monitored-agent",
signingKey: "your-hmac-key",
});
// Wrap tools for independent witnessing
const monitoredTool = witness.wrapTool(originalTool);
// Witness NHI lifecycle events discovered by Falcon Guardian
witness.witnessNhiLifecycle("service-account-47", "activated", {
scope: "read:customer-data",
ttl: 3600,
});
// npm install @tenova/swt3-mcp
// Add to your MCP client configuration:
{
"mcpServers": {
"swt3-witness": {
"command": "npx",
"args": ["-y", "@tenova/swt3-mcp"],
"env": {
"SWT3_API_KEY": "axm_live_...",
"SWT3_TENANT_ID": "YOUR_TENANT"
}
}
}
}
// 59 tools available. Use witness_tool_call, witness_nhi_lifecycle,
// witness_hardware, verify_trust, and more.
Every SWT3 procedure in this crosswalk maps to one or more regulatory frameworks. When Falcon Guardian detects and SWT3 attests, the resulting evidence satisfies requirements across:
| Framework | Relevant Articles / Controls | Falcon Guardian + SWT3 Coverage |
|---|---|---|
| EU AI Act | Art. 9 (risk management), Art. 11 (documentation), Art. 12 (record-keeping), Art. 14 (human oversight), Art. 15 (robustness) | Agent identity (Art. 12), tool witnessing (Art. 14), threat attestation (Art. 15), NHI lifecycle (Art. 9) |
| NIST AI RMF | GOVERN 1.2-1.7, MAP 1.1/5.2, MEASURE 2.6, MANAGE 2.3-2.4 | Trust verification (GOVERN 1.5), drift detection (MEASURE 2.6), access control (MANAGE 2.4) |
| OWASP Agentic Top 10 | All 10 risk categories (AGA-01 through AGA-10) | Full coverage via Guardian enforcement + SWT3 attestation |
| NIST 800-53 | AC-4, AU-2, IA-4, IA-5, SI-7, SI-10 | Access witnessing (AC-4), audit integrity (AU-2), identity (IA-4/5), input validation (SI-10) |
| CMMC v2.0 | AC.L2-3.1.1, AU.L2-3.3.1, IA.L2-3.5.1, SI.L2-3.14.1 | Access control, audit, identity, system integrity attestation |
| Five-Eyes Agentic AI | Tool Authorization, Chain-of-Custody, Identity Verification | AI-TOOL.1/2 (tool auth), AI-CHAIN.1/2 (custody), AI-ID.1 + AI-TRUST.1/2 (identity) |
When Falcon Guardian and SWT3 operate together: