
A structured look at how defenders and authorized security teams classify the first stage of intrusion chains, from phishing to supply chain, and why taxonomy discipline drives better detection engineering.
| Topic | Initial access tradecraft — the entry stage of intrusion chains |
| Category | Threat intelligence and detection engineering |
| Primary Use | Building shared vocabulary for MITRE ATT&CK-aligned detection, purple-team planning, and incident triage in authorized environments |
| Safe Use | Educational analysis for defenders and authorized security teams conducting detection engineering, tabletop exercises, and lab-based purple teaming |
| Telemetry Note | Each access vector maps to observable artifacts: authentication anomalies, endpoint execution traces, and mail-gateway logs that defenders should baseline |
Every intrusion has a beginning, and the industry has spent two decades arguing about how to name that beginning. Initial access tradecraft — the set of techniques an adversary uses to gain the first foothold in a target environment — is simultaneously the most and least understood phase of an intrusion chain. It is the most understood in the sense that virtually every incident response report opens with it, and the least understood in the sense that the vocabulary used to describe it varies wildly between vendors, government advisories, and academic taxonomies. This essay is an educational analysis for defenders and authorized security teams, mapping the major categorization schemes, examining where they agree and diverge, and explaining why taxonomy discipline is not pedantry but a prerequisite for measurable detection engineering.
The dominant public framework is MITRE ATT&CK, whose Initial Access tactic (TA0001 in the Enterprise matrix) organizes entry techniques into a small set of numbered families. Phishing appears in three flavors — spearphishing via attachment (T1566.001), via link (T1566.002), and via service (T1566.003) — reflecting the reality that delivery mechanism changes the telemetry significantly even when the social-engineering core is identical. Drive-by compromise (T1189), exploitation of public-facing applications (T1190), hardware additions (T1200), valid accounts (T1078), trusted relationships (T1199), and supply chain compromise (T1195) round out the headline categories. The framework's value is not that the list is exhaustive but that each technique is anchored to concrete, observable behavior, which lets authorized teams write detections against a stable identifier rather than a marketing label.
A parallel lineage of taxonomy comes from the vulnerability-management world, where initial access collapses into a single blunt question: is there an externally reachable, exploitable service? This framing powers scoring systems like CVSS, whose Attack Vector metric distinguishes network-adjacent, network-remote, local, and physical access. The disconnect between the two worlds is instructive. An attacker exploiting an internet-exposed appliance and an attacker logging in with purchased credentials both achieve initial access, but only the first shows up in a vulnerability-scanner-driven risk register. The surge in infostealer-derived credential markets over recent years has made the second path — valid accounts with no exploit, no malware on the perimeter, and often no alert — arguably the most operationally significant access vector in modern intrusions, and it is precisely the one that perimeter-centric taxonomies underweight.
Ransomware-focused reporting has produced a third, coarser vocabulary driven by leak-site attribution and affiliate structures. Initial access brokers — specialists who sell footholds rather than execute the full intrusion — have professionalized the entry stage to the point where taxonomy must account for an economic layer, not just a technical one. When analysts say an intrusion began with Fortinet VPN credential abuse versus Citrix NetScaler exploitation versus a compromised RMM tool account, they are really describing three different markets: credential resale, N-day exploitation of edge devices, and trusted-relationship abuse. Understanding the broker ecosystem explains why certain access vectors cluster around certain post-compromise behaviors, which in turn informs which detections deserve priority during triage of a confirmed foothold.
Supply chain compromise deserves separate treatment because it breaks the mental model most taxonomies assume: that the defender controls the boundary being crossed. Technique T1195 spans three sub-techniques — software supply chain compromise (T1195.002), compromise software dependencies (T1195.003 after SolarWinds-era refinement), and hardware supply chain compromise (T1195.001). The dependency variant is the hardest to operationalize defensively because the malicious code arrives through the same signed, trusted channel as legitimate updates, meaning traditional indicators like unsigned binaries or unusual download sources simply do not fire. Detection here shifts upstream to build pipelines, package registries, and behavioral analysis of what a legitimately delivered binary does after installation — a fundamentally different telemetry problem than perimeter exploitation.
Where taxonomies diverge most usefully is in their treatment of persistence-adjacent access. ATT&CK deliberately separates Initial Access from Persistence, but real intrusion chains frequently blur the line: an adversary who compromises a service provider under T1199 may hold standing access for months before the first detection-relevant action. Similarly, valid account usage (T1078) is simultaneously an access technique, a persistence mechanism, and a privilege-escalation enabler depending on when in the chain it is observed. Analysts who treat tactic assignment as a property of the adversary rather than a property of the observed event will misclassify these cases. The practical lesson for authorized purple teams is to map detections to behaviors and timestamps, not to the tidy narrative in an executive summary.
Detection engineering against each category follows distinct patterns worth spelling out. Phishing detections live at the mail gateway and at the moment of user interaction: attachment detonation telemetry, link reputation, and post-click endpoint execution tracing. Exploitation of public-facing applications (T1190) is best observed through the intersection of edge-device logs, WAF anomaly data, and process lineage on the exploited service itself — the classic signal being a web-worker process spawning an interactive shell or a scripting interpreter. Valid account abuse shows up not in malware telemetry at all but in authentication analytics: impossible travel, dormant-account revival, MFA fatigue patterns, and anomalous OAuth grant flows. Trusted relationship abuse appears in cross-tenant activity logs that many organizations do not centralize, which is why it persists as a blind spot years after major incidents demonstrated its effectiveness.
The taxonomy also matters for prioritization math. Not every initial access vector is equally available to equally capable adversaries. Exploitation of a public-facing application requires either an available exploit or the capability to develop one; credential abuse requires only money and a market. This asymmetry is why threat-modeling exercises that begin with the question "what does our exposure surface cost on the access-broker market?" often produce more honest risk assessments than those that begin with CVE counts. It also explains why hardened MFA deployment, continuous credential-leak monitoring, and aggressive deprovisioning of dormant accounts frequently outperform additional perimeter tooling in measured risk reduction — they attack the cheapest and most reliable access vector first.
For authorized security teams running exercises, a practical approach is to build a small internal matrix mapping each T1566/T1190/T1078/T1195/T1199 family to the specific telemetry sources available in their environment, and to mark honestly where no telemetry exists. Gaps discovered this way — a branch office's VPN without centralized logging, an RMM platform with no alerting on new device enrollment, a build server with unsigned artifact execution allowed — are more actionable than any generic maturity score. The exercise is defensive by construction: it never requires executing an attack, only reasoning about observable consequences.
A final observation on vocabulary hygiene: when incident reports, threat intel feeds, and detection rules use different names for the same entry behavior, correlation fails silently. A feed describing "credential replay," a SIEM rule matching "suspicious VPN login," and an ATT&CK-mapped analytics tagged T1078 are the same story told three ways. Teams that enforce a single canonical taxonomy in their detection-as-code repositories, and that require every analytic to cite a technique identifier, gain the ability to measure coverage empirically — which techniques have detections, which have telemetry but no logic, which have neither. That measurement loop, not the taxonomy itself, is the actual defensive product of all this categorization effort, and it is available to any authorized team willing to do the unglamorous bookkeeping.
Educational analysis for authorized security professionals. Use only in controlled, authorized environments.
Related coverage
0 comentários:
Post a Comment
Note: Only a member of this blog may post a comment.