k0nsult.cloud / ai-truth / ipIII / klasyfikacja / en

Classification Engine

Module 3 of the K0NSULT system. A deterministic engine that turns a raw report into a structured record across 6 independent axes. Classification is the input for risk scoring, playbook selection and computation of regulatory obligations — and every axis is auditable and versioned.

PL EN Intake Threat Map Incident Table Evidence Board Legal Board

DEMO MODE. All mappings and examples below are demonstrative. The regulatory frames (AI Act Art. 73, NIS2, GDPR Art. 33/34) rest on the publicly known text of those regulations and are marked as norm/frame — not as a description of a real incident.

6 classification axes + MITRE ATT&CK

The axes are orthogonal: the same incident type may involve different actors, vectors and impacts. Only the full set of axes yields a risk weight and a set of obligations.

1 layer2 type3 actor4 vector5 impact6 obligations+ ATT&CK

classification axes

orthogonal, auditable

technical layers

human → continuity

actor classes

cybercrime → agent

legal regimes

AI Act·NIS2·KSC·GDPR

Axis 1 — Technical layer

Where in the technology stack the incident plays out. It determines the lead team and the control set.

Layer	Scope	Typical incidents	Lead role
Human	social engineering, awareness	Phishing, vishing, BEC	SOC / awareness
Identity	accounts, MFA, tokens, sessions	Credential theft, session hijack	IAM / SOC
Application	web, API, mobile	Vulnerability, misconfig, injection	DevSecOps
Data	databases, backups, PII	Leak, ransomware, exfiltration	Legal/DPO + SOC
AI / agent	models, agents, plugins	Prompt injection, agent hijack, poisoning	AI Safety Officer
Infrastructure	network, cloud, endpoints	DDoS, malware, lateral movement	SOC / NetOps
Continuity	backup, DR, segmentation	Backup loss, DR failure	BCM / DevSecOps

Axis 2 — Incident type → default priority

Type	Layer	Default P	Suggested playbook
Ransomware	Data / infra	P0	playbook-ransomware
Data breach	Data	P0	playbook-data-breach
Agent hijack	AI / agent	P1	playbook-agent-hijack
Prompt injection	AI / agent	P1	playbook-prompt-injection
Phishing (active)	Human / identity	P1	playbook-phishing
DDoS	Infrastructure	P1	playbook-ddos
Exploited CVE	Application	P1	playbook-vulnerabilities
Supply chain	Application / infra	P2	playbook-supply-chain
Harmful hallucination	AI / agent	P2	playbook-hallucination
Misconfiguration	Application / infra	P3	playbooks

Axis 3 — Perpetrator / actor

Cybercrime

Financial, opportunistic motive. Mass scale.

RaaS

Ransomware-as-a-Service. Affiliates + operators, revenue-split model.

PhaaS

Phishing-as-a-Service. Ready-made kits, AiTM toolkits.

APT

Sponsored / persistent actor. Goal: long-term access, intelligence.

Hacktivist

Ideological motive. Often DDoS / defacement.

Insider

An internal person — malicious or accidental.

Operator error

Misconfiguration, faulty change, missing process.

Hostile AI agent

A hijacked or malicious agent acting autonomously.

Unknown

Default when there is no attribution — status GAP.

Attribution ≠ proof. The actor class remains a hypothesis until confirmed by an artifact. By default actor = unknown, status GAP. Do not promote to CONFIRMED on the basis of TTP alone.

Axis 4 — Entry vector → MITRE ATT&CK mapping

Vector	ATT&CK Tactic	Example Technique
Email (attachment/link)	Initial Access	T1566 Phishing
SMS (smishing)	Initial Access	T1566.002 Spearphishing Link
Website / watering hole	Initial Access	T1189 Drive-by Compromise
API	Initial Access / Credential Access	T1190 Exploit Public-Facing App
VPN / remote access	Initial Access	T1133 External Remote Services
Vulnerability (CVE)	Initial Access	T1190 Exploit Public-Facing App
Token / session	Credential Access	T1528 Steal Application Access Token
Prompt (AI input)	Initial Access (AI)	Prompt Injection (ATLAS AML.T0051)
Agent plugin / tool	Execution (AI)	Tool abuse / tool poisoning
Connector / integration	Lateral Movement	T1210 Exploit Remote Services
Dependency	Initial Access / Supply Chain	T1195 Supply Chain Compromise

AI vectors are auxiliary-mapped onto MITRE ATLAS (adversarial ML). The identifiers above are an example of a reference mapping, not an assignment to a real event.

Axis 5 — Impact

Confidentiality (C)

Data disclosure / exfiltration. Trigger for GDPR / banking secrecy.

Integrity (I)

Data modification, model poisoning, forged transaction.

Availability (A)

DDoS, ransomware, service outage. Trigger for NIS2 (service disruption).

Financial

Direct loss, fraud, recovery cost.

Legal

Notification obligations, penalties, supervisory proceedings.

Reputational

Loss of customer trust, media pressure.

Operational

Process interruption, team load, roll-back.

Physical

Impact on OT / critical infrastructure / personal safety.

Axis 6 — Regulatory obligations (norm / frame)

Classification condition	Flag	Obligation (regulatory frame)	Clock
Personal data compromised	GDPR_BREACH	GDPR Art. 33 — notification to the authority; Art. 34 — notification of individuals when high risk	72h to authority
Entity/service in NIS2 scope	NIS2_RELEVANT	NIS2 — early warning, notification, final report	24h / 72h / 1 month
KSC entity (PL)	KSC_RELEVANT	National Cybersecurity System — report to the competent CSIRT	per KSC
High-risk AI system	AI_HIGH_RISK	AI Act — requirements for Annex III systems	—
Serious AI incident	AI_SERIOUS_INCIDENT	AI Act Art. 73 — report a serious incident to the market surveillance authority	without undue delay
Critical infrastructure	CRITICAL_INFRA	Sectoral regimes + KSC; possible authority involvement	sectoral

Methodological note. The clocks above are a frame based on the public text of the regulations, not legal advice for a specific case. Final deadlines and addressees are confirmed by Legal/DPO on the Legal Board. Clocks run from the moment the entity became aware of the incident.

End-to-end classification example SIMULATION

Demonstration record — sample data, not operational.

{
  "id": "INC-DEMO-2026-0417",
  "evidence_status": "st-confirmed",       // artifact: AiTM proxy log
  "axis_1_layer":  "identity",
  "axis_2_type":   "phishing (AiTM)",
  "axis_3_actor":  "PhaaS",                 // hypothesis, TTP consistent
  "axis_4_vector": "email → website",
  "attack": { "tactic": "Initial Access", "technique": "T1566.002" },
  "axis_5_impact": ["C", "financial", "reputational"],
  "axis_6_legal":  ["GDPR_PERSONAL_DATA", "NIS2_RELEVANT"],
  "priority": "P1",                         // active phishing, sessions being hijacked
  "sla_h": 24,
  "playbook": "playbook-phishing"
}

CLASSIFICATION→RISK (P1)→PLAYBOOK phishing→OBLIGATIONS: GDPR 72h + NIS2 24h→REPORT

Consistency rule. The engine rejects a record if Axis 6 indicates a notification obligation while Axis 5 does not contain the corresponding impact (e.g. GDPR_BREACH without a "C" impact). Inconsistency → status DISPUTED and return to the Analyst.