Platform Architecture

What Makes a SOC Agentic?

"Agentic" is not a synonym for AI-powered. ARIA collects structured evidence deterministically before the AI agent runs, requires every verdict claim to cite that evidence, and verifies citations with code — not a second model guess — before any alert can be auto-closed.

AI-Powered vs Agentic — the Distinction Matters

Most security tools marketed as "AI-powered" apply a model to each alert independently — scoring it against a feature vector or pattern set and returning a number. That is AI applied to a security problem. An agentic system is different: it takes sequences of reasoning steps, uses tools to gather additional evidence, and adapts its investigation based on what it finds. The distinction is meaningful, not just terminological.

ARIA's pipeline is agentic in a specific sense. When an alert arrives, it does not immediately go to a language model for scoring. A deterministic enrichment stage runs first, querying multiple sources to build context: what MITRE technique the sensor tagged, whether the indicator appears in threat intelligence feeds, what this entity's recent alert history looks like, and — for cloud events — what identity context exists (MFA status, roles, recent login history). Only then does the LLM receive the alert, along with all gathered evidence.

The LLM produces a structured verdict. Crucially, each claim in the verdict must cite the specific query result that supports it — not infer from training data alone. A verdict that says "this IP appears in threat intelligence" must be supported by the threat intel lookup output in the enrichment trace. This citation requirement is enforced structurally, not by instruction alone.

Evidence-First Investigation — Structured Questions Before the Agent Runs

A key weakness of open-ended AI investigation is that the model decides what to look for. It may overlook a question that would change the verdict, not because the data is unavailable, but because it never occurred to ask. ARIA addresses this with a pre-collection phase that runs before the investigation agent gets the alert.

For each alert family, a defined set of investigation questions is answered deterministically using batched tool calls — without the AI. For an OAuth token event, for example, this covers whether the authorizing application has been seen before for this tenant, whether the authorizing user has seen this app, whether suspicious logins occurred in the same window, and whether co-occurring alerts suggest a broader pattern. These questions are answered before the agent's first turn, and the results are handed to the agent as pre-collected evidence.

Each evidence item carries an explicit status that distinguishes between meaningfully different outcomes. Confirmed means a search returned supporting data. No data means the search ran and returned zero results — absence of evidence, not evidence of absence. Tool error means data collection failed and the question is genuinely unanswered. Policy denied means the data source is disabled by the customer's own configuration. These distinctions matter: a verdict that says "no suspicious logins were observed" is only valid when the login search ran and returned zero results — not when the search failed or was blocked.

When the investigation agent receives the alert, it already has structured answers to the most important questions. Its tool budget is spent on investigation-specific queries, not re-asking questions the pre-collection phase already answered.

The Grounding Verifier — Code, Not a Second Guess

After the agent produces a verdict, a deterministic code check runs before auto-close is permitted. Every factual claim in the verdict that cites a pre-collected evidence reference is checked mechanically: does the cited evidence ID exist in this investigation's evidence set? Does it belong to the same tenant? Does its status actually permit the claim being made?

A verdict that says "the app was seen before by this user" and cites a pre-collected evidence object with status no data fails the grounding check — the cited evidence contradicts the claim, it does not support it. A verdict that cites a tool error evidence object as proof of absence also fails. These failures are caught by code before any auto-close decision, with no LLM judgment involved.

This is separate from the semantic verifier, which checks whether the verdict's reasoning is logically consistent. The grounding verifier checks a more basic question: are the citations real? An alert where the grounding check fails is routed to a security engineer regardless of how confident the triage model was.

The Semantic Verification Layer — a Second Model from a Different Vendor

Single-model AI systems have a correlated failure mode: if the model has a blind spot, it will fail consistently on the same class of input. ARIA addresses this by routing every verdict through a second model from a different vendor for independent semantic verification. The two models do not share weights, training data, or vendor infrastructure.

The verifier is explicitly adversarial: its task is to find reasons the verdict is wrong, not to confirm it. A verdict that the verifier cannot refute passes. A verdict where the verifier identifies a gap — missing evidence, a cited fact that does not match the source, a claim that exceeds what the data shows — is flagged as disputed. Disputed verdicts are blocked from both auto-close and client notification. They go to a security engineer.

This design means the system fails conservative rather than optimistic. When the two models disagree, a human sees the alert. When required enrichment data was not available, auto-close is blocked by the evidence-sufficiency check. The system's errors tend to create more work for the engineer, not fewer escalations — which is the right tradeoff for security.

What "Agentic" Does Not Mean Here

Agentic architecture does not imply autonomous action. ARIA's response capabilities are approval-gated: any action that would affect a client system — isolating a device, blocking an account, modifying a firewall rule — requires explicit sign-off from a security engineer before it executes. The agentic part of the pipeline stops at the routing decision.

This is a deliberate constraint, not a limitation of the technology. Autonomous containment actions can have operational consequences — blocking a legitimate account, isolating a production server — that are sometimes worse than the threat they respond to. Until the confidence bar for fully autonomous response is higher than current AI systems can reliably clear, requiring human approval before response is the correct architecture.

The agentic layer handles the part of security work that doesn't require operational judgment: gathering evidence, correlating context, reasoning about whether an alert represents a real threat, and routing appropriately. The human handles the part that does: deciding whether to contain, what to contain, and when.

What the agentic pipeline covers:

  • Deterministic enrichment across MITRE, threat intel, and alert history
  • Evidence pre-collection: structured investigation questions answered before the AI agent runs, with explicit status per question (confirmed / no data / tool error / policy denied)
  • Structured LLM investigation with source-cited verdicts, guided by pre-collected evidence
  • Code-level grounding verifier: mechanically checks that verdict citations point to real, non-absent evidence — no LLM judgment
  • Semantic verification via a second model from a different vendor
  • Evidence-sufficiency check blocking auto-close when required data was unavailable or failed
  • Routing: benign auto-close, hold for engineer, or escalation
Common Questions
What is the difference between an AI-powered SOC and an agentic SOC?
An AI-powered SOC applies a model to each alert — typically scoring it or classifying it in a single pass. An agentic SOC takes multiple investigation steps: querying evidence sources, reasoning across what it finds, verifying conclusions, and routing based on confidence. ARIA's pipeline does not hand an alert to a model and accept its output. It runs enrichment first, requires the model to cite evidence in its verdict, and passes the verdict through independent verification before any automated action.
Does ARIA's agentic pipeline take action autonomously?
No. The pipeline handles investigation and routing — it does not execute response actions without approval. Any action affecting a client environment (device isolation, account suspension, firewall changes) requires explicit sign-off from a security engineer. The agentic reasoning stops at the decision; the execution is always gated.
What happens when the verifier and triage model disagree?
The verdict is flagged as disputed and held for a security engineer. It is not auto-closed and not sent to the client. The engineer sees both the triage verdict and the verifier's objection and can review the underlying enrichment data before deciding. The default is conservative: when in doubt, a human decides.
What is the evidence-sufficiency check?
Before any alert can be auto-closed, ARIA checks whether the required enrichment sources were actually consulted. If a required lookup failed, timed out, or returned no data, the sufficiency check blocks auto-close regardless of what the triage model concluded. An alert can only be automatically resolved when the evidence base that the verdict rests on is confirmed to be complete.
What is the evidence pre-collection phase?
For supported alert families — starting with OAuth and cloud identity events — ARIA runs a set of structured investigation questions before the AI agent sees the alert. These questions are answered with batched, deterministic tool calls: has this app been seen before for this tenant? Did suspicious logins occur in the same time window? Are there co-occurring alerts? The answers are handed to the agent as pre-collected evidence with an explicit status per question. This means the agent starts its investigation with grounded data rather than an open field, and its tool budget is spent on investigation-specific queries rather than re-asking questions that can be answered deterministically.
Why does ARIA distinguish between "no data" and "tool error"?
These are meaningfully different outcomes. "No data" means the search ran successfully and returned zero results — that is evidence of absence. "Tool error" or "source unavailable" means the search failed and the question is genuinely unanswered. A verdict that concludes "no suspicious logins were observed" is only valid when the login search ran and returned zero results. If the search failed, the agent has no basis for that conclusion. ARIA tracks nine evidence statuses — confirmed, contradicted, no data, tool error, timeout, policy denied, source unavailable, not applicable, and insufficient history — and the grounding verifier enforces that verdicts only cite statuses that actually support the claim being made.
What is the grounding verifier?
The grounding verifier is a deterministic code check — not an LLM — that runs after the triage agent produces a verdict and before any auto-close decision. It checks every evidence reference in the verdict: does the cited evidence ID exist in this investigation's collected evidence? Does it belong to the same tenant? Does its status permit the claim? A verdict that cites a "no data" evidence object as positive proof fails the check. A verdict that cites a "tool error" object as proof of absence also fails. Failures block auto-close and route the alert to a security engineer. This check runs before the semantic verifier, which evaluates whether the verdict's reasoning is logically sound.
Does the agentic pipeline apply to every alert?
Not entirely. Known-benign patterns — specific process-signature combinations that reliably produce false positives — are closed by deterministic rules without invoking AI. These bypass rules are narrow and maintained explicitly. Every other alert goes through the full enrichment and triage pipeline. Evidence pre-collection applies to supported alert families (currently OAuth and cloud identity events); other alert types go through the standard enrichment and investigation flow.

See ARIA in Practice

Book a free assessment to see how the pipeline handles your actual alert environment.

Book Free Assessment
No contract. No setup fee. Cancel anytime.

See Plans

Starting at $799/month. No long-term contracts required.

View Pricing