Intrusions investigated by Mandiant during 2025 had a global median dwell time of 14 days, up from 11 days in 2024, according to M-Trends 2026. For a security operations center (SOC), the practical challenge is recognizing which observations reveal an intrusion and acting before the attacker progresses.
Attack signal intelligence connects security observations with attacker behavior, affected identities and systems, and operational context. Its purpose is to give defenders a prioritized, evidence-backed account of suspected activity that they can investigate and act on.
What is attack signal intelligence?
Attack signal intelligence connects security observations, affected entities, and attack context to identify and prioritize suspected attacks. A useful signal explains what is suspicious, who or what is involved, why the activity matters, and which evidence an analyst should examine.
Vectra AI uses Attack Signal Intelligence (ASI) for its AI-driven detection, triage, and prioritization capability. Its attack signal intelligence capability includes correlating detections, attributing activity, prioritizing entities, and constructing attack narratives.
The vocabulary matters because each layer represents a different level of interpretation:
- Event: An observed action, such as a sign-in or an administrative change.
- Alert: A notification that a detection condition has been met.
- Signal: Relevant evidence interpreted in context, including the behavior, affected entity, and reasons for concern.
- Entity: An account, device, workload, application, or other identifiable participant in the activity.
- Incident: A security event or related set of events that the organization manages through its incident response process.
These are working distinctions for the workflow described here. An alert can contain valuable signal, and a single high-confidence observation can justify immediate investigation. Correlation should add context without delaying action that the evidence already supports.
Why connected attack signals matter?
Connected attack signals help security teams recognize attacks that use legitimate credentials and administrative functions across different systems. An action's technical validity does not establish its business legitimacy, so analysts need context to distinguish expected work from potential compromise.
Adversaries can abuse existing accounts without deploying malware, a behavior documented in MITRE ATT&CK's Valid Accounts technique. A successful sign-in may therefore be the beginning of an investigation rather than evidence that the activity is safe.
Consider an account that authenticates from a new device, accesses an unfamiliar server, and then changes a cloud permission. Each action may have an innocent explanation. The relationship between the account, device, sequence, target, and expected duties determines whether the activity deserves urgent attention. This is where understanding attacker behavior becomes useful.
Current incident research also shows why security teams need more than one success measure. The same M-Trends 2026 research reports that internal discovery increased from 43% of investigated incidents in 2024 to 52% in 2025, even as median dwell time increased. The research draws on more than 500,000 investigative hours globally in 2025. Its engagement sample and changing incident mix limit broad generalization.
The operational inference is that a higher share of internally discovered incidents can coexist with longer undetected intrusions. Neither measure establishes the effectiveness of a particular detection product. A SOC should examine coverage, discovery speed, investigation quality, and missed behavior together.
How attack signal intelligence works?
Attack signal intelligence turns telemetry into prioritized investigations through collection, behavioral detection, entity attribution, triage, and investigation support. Implementations may combine these tasks differently; the evaluation question is whether the resulting evidence supports a sound decision.
Collect telemetry with enough context
Collect telemetry that can identify the participating entities and reconstruct the behaviors your team needs to investigate. More data is useful when it closes a meaningful visibility gap or preserves information needed to reconstruct activity.
| Evidence source |
Useful observations |
Investigation question |
| Identity and access systems |
Authentication, session, device, role, and privilege changes |
Who or what obtained access, and under which conditions? |
| Network activity |
Connections, destinations, protocols, timing, and traffic characteristics |
Which systems communicated, and how did access expand? |
| Cloud and SaaS audit logs |
Administrative actions, API activity, application access, and data operations |
What changed or was accessed after authentication? |
| Endpoint and workload telemetry |
Processes, remote sessions, execution, and local changes |
What occurred on the participating system? |
| Asset and business context |
Owners, privileges, application dependencies, and criticality |
What would compromise of this entity affect? |
Preserve identifiers that connect these observations, including account, device, session, resource, and event identifiers where available. Record event time and collection time separately so delayed ingestion does not appear to be delayed attacker activity.
Build collection around useful event detail, consistent timestamps, centralized correlation, and protected logs, as recommended in the joint government guidance on event logging and threat detection. Build collection and retention decisions around the investigations you need to support, with explicit attention to sensitive data and operational constraints.
Track collection failures and gaps alongside detections. An unavailable audit feed reduces what the team can conclude about the environment. For a deeper foundation, start with the security telemetry your investigations need.
Detect behaviors in their operating context
Behavioral threat detection examines actions and relationships that may indicate an attack. Examples include unexpected use of privileged access, unusual remote administration, discovery activity, or access patterns inconsistent with an entity's role.
An anomaly establishes a deviation from a baseline. Assessing maliciousness requires additional reasoning about the action, permissions, timing, target, and surrounding evidence. A new administrator performing approved maintenance may produce unusual activity that is entirely legitimate.
Signatures, rules, known indicators, and behavioral models can contribute different evidence. Ask which behaviors an implementation detects, which data it requires, how environmental changes affect it, and how analysts can review the explanation. The presence of AI does not answer those questions by itself.
Connect activity to the right entities
Correlation groups observations that have a meaningful relationship. Entity attribution connects that activity to the account, device, workload, or application involved. Here, attribution means identifying participants in the environment; naming a threat actor requires a separate evidentiary basis.
This step needs care. An IP address may represent a shared gateway, several users, or different devices over time. A shared account may conceal the individual operator. A short-lived workload may disappear before an analyst begins investigating.
Prefer stable identifiers and corroborating evidence when available. Preserve uncertainty when the relationship is ambiguous. Two alerts generated from the same underlying event also provide less independent corroboration than observations from separate sources.
During incident investigation, analysts need the activity sequence and the basis for each connection. Analysts should be able to ask whether two entities were correctly linked and inspect the source observations behind that relationship.
Triage evidence and prioritize the response
Triage assesses the observations and their plausible explanations. Prioritization determines which investigation deserves attention first as evidence and business context change.
Keep these considerations distinct:
- Confidence: How strongly does the available evidence support the suspected behavior?
- Severity: How consequential could the behavior be if malicious?
- Business importance: Which services, privileges, or sensitive resources could be affected?
- Urgency: How quickly does the team need to investigate or respond given the present evidence?
A suspected compromise involving a privileged identity may warrant prompt investigation even while some details remain uncertain. Repeated low-impact observations on a test system may deserve a different response. Ask how a priority changes when new activity arrives or an entity's business role changes.
Treat a score as a prioritization aid unless its calibration supports a stronger interpretation. A score of 90 should not be read as a 90% probability of compromise without evidence that the score has that meaning.
Reducing repetitive notifications can help limit alert fatigue and keep analysts focused. Retain the underlying observations when they may become useful later, and document exceptions so legitimate maintenance does not create an indefinite blind spot.
Support investigation and controlled action
A prioritized signal should help an analyst test a hypothesis and decide what to do next. Useful investigation context includes the entities involved, observed actions, sequence, source records, reasons for priority, possible impact, and unresolved questions.
Separate observed facts from inferred relationships and recommended next steps. If an AI-generated summary says an account accessed a sensitive resource, an analyst should be able to inspect the supporting record and its timestamp.
Execution is a separate responsibility. Disabling an account, isolating a device, or changing a network policy requires supported controls, sufficient permissions, and a response process appropriate to the service affected. Verify which actions a deployment can recommend, request, approve, execute, and confirm.
The output may feed an analyst console, ticketing workflow, security information and event management (SIEM) platform, or security orchestration, automation, and response (SOAR) system. The useful outcome is a traceable handoff with clear ownership and action status.
Attack signal intelligence and related security approaches
Attack signal intelligence connects and prioritizes evidence about suspected activity, while threat intelligence supplies broader context and security platforms provide collection, analysis, or response functions. These roles can overlap within a deployment.
| Approach |
Main contribution to the workflow |
What to examine |
| Threat intelligence |
Context about threats, behaviors, infrastructure, and other relevant observations |
Whether the information applies to the observed activity and has adequate provenance |
| Anomaly detection |
Identification of deviations from expected patterns |
Whether the deviation is meaningful, explainable, and supported by other context |
| SIEM and extended detection and response (XDR) |
Collection, analysis, correlation, and investigation capabilities, depending on implementation |
Actual source coverage, entity relationships, retained evidence, and workflow fit |
| SOAR and response integrations |
Coordination or execution of actions and playbooks |
Permissions, approval boundaries, error handling, and verification of the resulting action |
Threat intelligence provides processed or enriched information to support decisions, consistent with NIST's definition. Threat intelligence extends beyond lists of known malicious indicators. It can inform behavioral detection, investigation hypotheses, and interpretation of activity inside an organization.
The practical distinction is the question being answered. Threat intelligence can explain relevant threats and methods. Attack signal analysis applies observations and context to suspected activity in the monitored environment. The two can reinforce each other. For more on the intelligence workflow, see AI in threat intelligence.
Evaluate the deployed capabilities rather than assuming a category label guarantees the required result. Existing platforms may already perform parts of this workflow; identify the remaining gap before adding another component.
An example of an attack signal investigation
An attack signal investigation can connect a suspicious sign-in, unusual remote access, and a cloud permission change into a testable account-compromise hypothesis. This hypothetical example shows how the observations can change an investigation's priority while leaving room for legitimate explanations.
The scenario draws on the behaviors described in MITRE ATT&CK's Valid Accounts and Remote Services techniques. It represents a defensive evaluation example, not a documented customer incident or a claim that a product detects every step.
| Observation |
Why it could matter |
Evidence to seek next |
| A privileged account signs in from an unfamiliar device |
The account may be compromised, or an administrator may have replaced a device |
Device identity, session details, access policy, and approved changes |
| The account uses remote services to reach an unusual internal server |
The activity may indicate lateral movement or legitimate support |
Source and target identifiers, remote-session records, and the account's expected duties |
| A related session changes access to a sensitive cloud resource |
The sequence may expand access or reflect an authorized deployment |
Session linkage, audit events, resource ownership, and the permission change |
| A workload begins transferring data to a new external destination |
The activity may indicate exfiltration or an approved backup process |
Process or workload context, destination ownership, transfer characteristics, and change records |
The investigator can now examine a connected hypothesis: a potentially compromised identity used access across several systems and reached sensitive resources. The narrative should show which connections are established and which remain uncertain.
Suppose the cloud event lacks a session identifier and the internal access passed through a shared gateway. Preserve those limits. The evidence may support several related investigations while the team checks whether the activity belongs to the same operator.
Disconfirming evidence matters too. An approved maintenance window, verified device replacement, and expected backup destination could explain parts of the sequence. A sound workflow supports revising or closing the hypothesis while retaining the rationale and relevant records.
How to evaluate attack signal intelligence?
Evaluate attack signal intelligence by testing whether analysts can identify affected entities, inspect supporting evidence, and reach correct decisions within useful timeframes. A useful proof of value measures investigation quality, coverage, and time alongside the number of alerts or cases requiring review.
Use realistic attack and benign scenarios
Choose a small set of approved defensive exercises that reflect your environment. Include a sequence crossing identity and network boundaries, a cloud or SaaS action relevant to your business, and legitimate administration that could resemble those behaviors.
For each exercise, agree on the starting conditions, available telemetry, expected observations, meaningful entity relationships, and acceptable response boundaries. Record the software version, enabled integrations, tuning, and scope so another reviewer can understand the result.
Include a degraded-data condition, such as a delayed audit feed or missing identifier. Examine whether the system reports the gap, preserves uncertainty, and still presents the evidence it does have. Use safe exercises within the organization's authorization and operating constraints.
Review the investigation with the analyst who would actually handle it. Can that person explain why the entity was prioritized, locate the source evidence, identify missing context, and decide on an appropriate next action?
Measure quality at the same unit of work
Measure detection quality and effort using an explicit unit of work, such as events, alerts, prioritized entities, investigation cases, or incidents. Aggregating many alerts into one case can reduce the review queue without changing the number of underlying detections or their correctness.
| Measure |
Practical definition for the evaluation |
Interpretation limit |
| Reviewed-case precision |
Confirmed malicious cases divided by reviewed cases with a resolved verdict |
Report unresolved and unreviewed cases separately; review selection can bias the result |
| Exercised-behavior detection |
In-scope exercised behaviors detected divided by in-scope behaviors exercised |
A controlled-test measure depends on scenario scope and does not establish production-wide recall |
| Time to useful prioritization |
Time from the first relevant observed event to an investigation an analyst can use |
Separate collection delay, detection delay, and prioritization delay |
| Investigation effort |
Analyst time needed to reach a documented disposition |
Hold case complexity and the stopping rule reasonably consistent |
| Attribution and evidence quality |
Correct entity links, accessible source records, and visible gaps in the exercised cases |
Requires inspection of the underlying records, not only the summary |
Track distributions or slow cases where they matter; a single average can hide delayed investigations. Measure containment time separately from detection and triage so integration or approval delays remain visible. When reporting mean time to respond (MTTR), document the interval from a confirmed or high-confidence detection to verified containment.
Lower alert volume is useful evidence about workload. Establishing better detection quality also requires examining what was detected, what was missed, and whether analysts could make correct decisions. Publish any claimed improvement with its baseline, period, denominator, and deployment conditions.
Ask for evidence behind each capability
Evaluate each claimed capability against documented source coverage, evidence access, response permissions, and the actual deployment scope:
- Which environments and entities are covered by the proposed deployment, and which remain outside it?\
- What data sources, permissions, collection methods, and integration prerequisites are required?
- How are entity relationships established, and how are uncertain relationships presented?
- Can analysts inspect or export source evidence, timelines, priority explanations, and dispositions?
- Where is data processed, how long is it retained, and how are access and auditability managed?
- Which response actions are available in the proposed package, and what approvals and controls do they require?
- What tuning, maintenance, integration work, services, and contractual allowances affect the total deployment cost?
Where a supplier cannot substantiate a capability in your scope, record the evidence gap and its operational consequence. The right result may be a targeted integration, a collection improvement, or a platform change, depending on the workflow your team needs.
Putting the workflow into security operations
Introduce attack signal intelligence into SOC operations through one defined investigation, with agreed evidence requirements, an owner, and a response path. Expand after the team can validate case handoffs and review outcomes.
- Choose the investigation. Define the suspected behavior, participating domains, affected business service, and decision the analyst must make.
- Confirm the evidence. Check collection health, timestamps, identifiers, permissions, and retention against that investigation's needs.
- Assign ownership. Identify who reviews a prioritized case, who can provide missing context, and who authorizes consequential actions.
- Connect the handoff. Verify how evidence, priority changes, case status, and executed actions move between the detection platform and existing workflows.
- Review outcomes. Examine malicious, benign, unresolved, and missed exercise cases. Use the findings to improve collection, attribution, detection engineering, and analyst guidance.
Keep exceptions specific and reviewable. Document the reason for a suppression, its scope, and when it should be revisited. Maintain a way to investigate retained observations when new evidence changes their significance.
As the workflow matures, expand to another relevant domain or attack path. Tie progress to the quality and timeliness of the resulting investigations, using threat detection as the broader operational context.
Explore the AI behind the attack signal
Discover how Vectra AI combines security research, behavioral detections, and AI agents to identify attacker activity and prioritize urgent threats.
Explore Vectra AI’s approach →
How Vectra AI approaches attack signal intelligence?
Vectra AI positions attack signal intelligence alongside unified observability, behavioral AI, and operational control within its AI-native security and observability approach. The relevant goal is to connect visibility and detection with a prioritized understanding of suspected activity and a clear response path.
Vectra AI describes its ASI capability as correlating detections, attributing activity, prioritizing entities, and constructing attack narratives. Assess those capabilities against the sources, environments, integrations, and investigation requirements in your evaluation.
For a customer perspective, True Digital Group's prioritization example describes using ASI to focus on critical threats. The public customer page provides a qualitative account rather than a transferable performance benchmark.
To explore how this applies to an existing SOC workflow, explore the threat detection and response use case. If you are ready to evaluate the approach, request a Vectra AI demo using your investigation scenario.
Note: This guide draws on public technical guidance and incident research. The investigation scenario and evaluation method are proposed guidance.