Agentic Security Explorer (ASE)

Learn Agentic Attack Surfaces, Risks, and Business Impact

Business Guide - Components

Component Guide

Each card explains what the component is, why it matters, what can go wrong, and the business impacts to consider.

OverviewLayersComponentsBusiness Impacts
Central Planner

Orchestrator

Breaks the top-level goal into subtasks and decides which agent or tool handles each one.

Why it matters: Orchestrator helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Goal manipulation via prompt injection
  • Single point of failure
  • Insecure delegation logic
Task Workers

Sub-Agents

Specialized agents that carry out the subtasks the orchestrator assigns.

Why it matters: Sub-Agents helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Excessive agency
  • Role confusion / impersonation
  • Insecure inter-agent trust
Function Execution

Tool-Calling Layer

The interface agents use to call external functions, APIs, and code.

Why it matters: Tool-Calling Layer helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Tool poisoning
  • Arbitrary code execution
  • Unvalidated trust in tool output
State & Recall

Memory / Context Store

Persists conversation history, long-term memory, and embeddings across turns and sessions.

Why it matters: Memory / Context Store helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Memory poisoning
  • Cross-session data leakage
  • Persistent prompt injection
Message Bus

Inter-Agent Comms

The channel agents use to pass messages, requests, and results to one another.

Why it matters: Inter-Agent Comms helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossLimited business impactMonitoring gapOperational disruptionOperational impactProcess inefficiencyRegulatory exposureReputation damage

What could go wrong

  • Man-in-the-middle injection
  • Message spoofing
  • Lack of message integrity checks
Knowledge Grounding

Retrieval / RAG

Fetches external documents or data to ground agent responses in current information.

Why it matters: Retrieval / RAG helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossLimited business impactMonitoring gapOperational disruptionOperational impactProcess inefficiencyRegulatory exposureReputation damage

What could go wrong

  • Indirect prompt injection via retrieved content
  • Data poisoning of the knowledge base
  • Source spoofing
Runtime Isolation

Execution Sandbox

Isolates code execution, browsing, and file operations from the host and from other agents.

Why it matters: Execution Sandbox helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Sandbox escape
  • Privilege escalation
  • Insufficient isolation between agents
Third-Party APIs

External Integrations

Connections to outside services — email, calendars, payments, ticketing, and more.

Why it matters: External Integrations helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Data exfiltration via a legitimate channel
  • Over-permissioned API scopes
  • Third-party supply chain compromise
Auth Material

Credentials & Secrets

API keys, tokens, and service credentials that let agents authenticate to tools and integrations.

Why it matters: Credentials & Secrets helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Secret leakage via logs or output
  • Long-lived static credentials
  • Credential reuse across agents
Approval Gate

Human-in-the-Loop

Checkpoints where a person reviews or approves an agent's proposed action before it executes.

Why it matters: Human-in-the-Loop helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossLimited business impactMonitoring gapOperational disruptionOperational impactProcess inefficiencyRegulatory exposureReputation damage

What could go wrong

  • Bypassable confirmation step
  • Approval fatigue / rubber-stamping
  • Unclear accountability
Authority Plane

Agent Identity & AuthZ

Issues, verifies, scopes, and revokes agent identities and delegated authority for actions.

Why it matters: Agent Identity & AuthZ helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Rogue agent registration
  • Broken delegated authorization
  • Stale or revoked token reuse
Behavior Packages

Agent Skills / MCP Registry

Reusable skills, MCP servers, manifests, and workflow packages that define what agents can do and how they do it.

Why it matters: Agent Skills / MCP Registry helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Malicious skill or MCP package installation
  • Skill typosquatting / brand impersonation
  • Unsigned or unpinned skill update
Instruction/Data Split

Context Boundary & Provenance

Separates instructions from data and preserves evidence, source, and trust level as context moves across agents and tools.

Why it matters: Context Boundary & Provenance helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Context boundary collapse
  • Provenance loss across hops
  • Evidence laundering through summaries
Reasoning Runtime

Model / Inference Routing

Selects model, prompt template, policy tier, context window, and fallback behavior for agent reasoning.

Why it matters: Model / Inference Routing helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossLimited business impactMonitoring gapOperational disruptionOperational impactProcess inefficiencyRegulatory exposureReputation damage

What could go wrong

  • Prompt template injection
  • Unsafe model fallback
  • Unbounded reasoning / denial of wallet
Outbound Surface

Output & Egress Channels

Represents every route through which the agent sends content, files, actions, records, or instructions outward.

Why it matters: Output & Egress Channels helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Covert exfiltration through normal output
  • Malicious generated artifact
  • Downstream automation abuse
TrustOps

Agent Lifecycle & Governance

Tracks agent inventory, owners, versions, approvals, deployment state, rollback paths, and emergency disablement.

Why it matters: Agent Lifecycle & Governance helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossLimited business impactMonitoring gapOperational disruptionOperational impactProcess inefficiencyRegulatory exposureReputation damage

What could go wrong

  • Missing kill switch or rollback path
  • Shadow agents outside governance
  • Untracked agent version drift
Forensic Plane

Telemetry Integrity & Evidence

Captures tamper-resistant evidence of plans, prompts, retrieved context, tool calls, approvals, outputs, and policy decisions.

Why it matters: Telemetry Integrity & Evidence helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossLimited business impactMonitoring gapOperational disruptionOperational impactProcess inefficiencyRegulatory exposureReputation damage

What could go wrong

  • Agent-generated telemetry tampering
  • Incomplete replayability
  • Non-repudiation failure
Policy & Observability

Guardrails & Monitoring

Policy enforcement, logging, and anomaly detection layered across the whole system.

Why it matters: Guardrails & Monitoring helps the organization use agentic AI capabilities while preserving continuity, control, and accountability.

Business value

  • Faster workflows
  • Reduced manual effort
  • Improved continuity
  • Scalable decision support

Business impact

Control weaknessData exposureFinancial lossOperational disruptionOperational impactRegulatory exposureReputation damage

What could go wrong

  • Guardrail bypass via obfuscation
  • Blind spots / incomplete logging
  • Telemetry tampering / false audit trail