← Blog
47 min readOlesia Shelestova

NIS2: Who Actually Has to Comply, What It Costs to Ignore and Why Nmap/AI Can't Prove It for You

A sourced walk through NIS2: who is in scope and what the fines really are, whether you need a certified auditor, and why the wave of nmap/Kali/AI 'NIS2 checkers' can't produce compliance evidence - with a full, control-by-control table of all 226 requirements mapped against Pentesterra, Nmap+NSE and AI-wrapped Kali.

NIS2complianceEU regulationpentestresearchGRC

Half the questions I get about NIS2 are some version of "do we actually have to do this, or is it another directive that sits in a drawer?"

Short answer: you have to do it, the drawer option expired, and the tools being sold to make it painless mostly can't.

Let's take that apart, with sources, and end somewhere useful.

NIS2 is Directive (EU) 2022/2555. A directive is not a regulation, so it does not bite you directly. It binds you through your own country's law, the way GDPR does. That single fact is behind most of the confusion, and it is where we should start, because "for whom is this mandatory" has a precise answer and "what happens if you ignore it" has numbers.

I've spent the last stretch reading the actual legal text, the Commission's implementing regulation, and the national transpositions, and then mapping all of it against what security tools can and can't do. This post is that work, in four parts, and then an honest pitch for where Pentesterra fits. Every claim links to a source. I would rather you check them.

Part 1: Who has to comply, and what it costs to ignore

The two-part test

You are in scope if both are true:

  • you operate in one of NIS2's 18 sectors, and
  • you are at least a medium-sized company.

The sectors come from two annexes:

  • Annex I (high criticality): energy, transport, banking, financial market infrastructure, health, drinking water, waste water, digital infrastructure, ICT service management (managed service and managed security providers), public administration, space.
  • Annex II (other critical): postal and courier, waste management, chemicals, food, manufacturing (medical devices, electronics, machinery, vehicles), digital providers (marketplaces, search engines, social networks) and research.

"Medium-sized" follows the EU SME definition: from 50 staff, or annual turnover and balance sheet both over 10 million euro. Below that you are usually out, with exceptions that ignore size entirely:

  • DNS providers, TLD registries, trust service providers and public electronic communications;
  • anyone a member state names as sole or critical provider (NIS2 Art.2 and Art.3);
  • if you sell across the EU but sit outside it, you still need an EU representative (Art.26).

Essential vs important is not a compliment, it's your supervision regime

NIS2 splits in-scope entities into essential and important. It is not a ranking of how good your security is. It decides how the regulator watches you and how hard they can hit you.

10M€ / 2% Max fine for essential entities: at least 10 million euro or 2% of worldwide turnover, whichever is higher Directive (EU) 2022/2555, Art.34(4)
7M€ / 1.4% Max fine for important entities: at least 7 million euro or 1.4% of worldwide turnover, whichever is higher Directive (EU) 2022/2555, Art.34(5)
CEO For essential entities a regulator can temporarily ban the chief executive from managing the company until it complies Directive (EU) 2022/2555, Art.32(5)

Essential entities get proactive supervision: on-site inspections, random checks, regular and targeted audits, and, in the text itself, "security scans" (Art.32(2)). Important entities get reactive supervision, triggered when there is an indication something is wrong (Art.33). Either way, management bodies must approve the security measures, oversee them, and can be held personally liable (Art.20). That last part is why boards started paying attention.

So has anyone actually been fined?

Here is the honest version, because you'll see scarier claims. There is not yet a long public list of NIS2 fines. The reason is timing: the transposition deadline was 17 October 2024, and most countries missed it. In July 2026 the Commission took Ireland, Spain, France and the Netherlands to the EU Court of Justice for still not having transposed, asking for lump-sum and daily penalties. So the enforcement machine is only now switching on.

What is real and dated is the exposure. Germany's law (NIS2UmsuCG) came into force in December 2025, and the number of supervised entities jumped from about 4,500 to roughly 29,500, with a registration deadline in March 2026. Belgium requires its essential entities to have a conformity assessment under way by 18 April 2026. Italy's ACN set significant-incident reporting from 1 January 2026 and baseline measures for most listed entities by around 31 October 2026. The fines are not hypothetical, they are scheduled.

~29,500 German entities now under NIS2 supervision, up from ~4,500 under the old KRITIS regime BSI, NIS2UmsuCG in force Dec 2025
4 Member states referred to the EU Court of Justice in July 2026 for failing to transpose NIS2 European Commission, IP/26/1499
24h / 72h Early warning within 24 hours, full notification within 72 hours, final report within one month of a significant incident Directive (EU) 2022/2555, Art.23

The takeaway for Part 1: if you are a medium or larger company in one of the 18 sectors, this is mandatory now, the penalties reach into eight figures and into your executives personally, and the countries that were slow are being pushed by the Court. The drawer is closed.

Part 2: Do you need a certified auditor to "get NIS2 certified"?

Short version: there is no EU "NIS2 certificate." NIS2 works like GDPR: you have to be compliant and be able to show it when the authority asks, and nobody hands you a certificate that ends the conversation. This is in the text, not my reading of it. The directive makes cybersecurity certification explicitly voluntary - a member state may require you to use certified products or services under a European scheme, but there is no "certified NIS2" status for the entity (Directive Art.24). What actually binds you is the duty to produce "evidence of implementation ... and the respective underlying evidence" when the authority asks (Art.32(2)(g), 33(2)(f)) - the same be-compliant-and-demonstrate-it model as GDPR's accountability principle.

A few things follow from that, and they are worth getting right because vendors blur them.

  • The Commission implementing regulation (CIR 2024/2690) spells out the technical and methodological measures in a 13-section annex of roughly 160 numbered requirements. It is directly binding on digital-infrastructure providers (DNS, cloud, data centres, MSPs, and so on). For everyone else the detailed measures come from national law, but the CIR annex, mirrored by ENISA's June 2025 implementation guidance, is the practical reference. That is the structure this whole analysis is built on.
  • Who checks you is your national authority: ACN in Italy, BSI in Germany, CCB in Belgium. Essential entities are checked proactively, important ones after a signal (Art.32 and 33). Targeted audits by an independent body are paid by the audited company (Art.32(2), 33(2)). Read that twice: the audit is on your bill.
  • "Certified auditor" is not the same as "certified you." Expect vendors to wave credentials around, so separate two things they blur. A person or firm can hold audit qualifications (CISA, ISO 27001 Lead Auditor) or be an accredited conformity-assessment body, and for a targeted audit NIS2 does require a qualified independent body (Art.32(2)(e)) - whose bill you pay. None of that changes what the tool in the engagement produces: no credential turns a banner-grabbing scan into reproducible evidence, and no private auditor issues a state "NIS2 certificate" - the national authority supervises, and the auditor's report is one input to it. So when a service is sold as "assessment by certified auditors," the question that matters is what evidence the assessment actually leaves behind, not whose signature is on the cover.
  • ISO 27001 is not automatically NIS2. It covers most of the Art.21 areas and is a strong evidence base, but it does not cover the Art.23 reporting timelines, national registration, or the specific management-liability duties. Belgium formally accepts ISO 27001 (with the right scope and Statement of Applicability) or its own CyberFundamentals framework as a route. Germany only forces recurring proof (every three years, via audit or certification) on critical-infrastructure operators, not on every entity.
  • Something is coming, but it is not here. The direction of travel is simplification and certification. The Commission's Digital Omnibus (19 November 2025) already unified incident reporting and clarified how NIS2 sits next to the CER Directive and DORA. Then on 20 January 2026 it proposed a full cybersecurity package: targeted NIS2 amendments plus a revised Cybersecurity Act. The headline for compliance is a cyber-posture certificate - the Commission wants entities to "certify their cyber posture ... [and] get presumption of conformity with NIS2 and other Union legislations." An essential entity holding one could be spared targeted audits for the covered requirements, but only where the certificate comes from a third-party conformity assessment body (Art.78(2) of the revised Act), and "on-site inspections, security scans, or targeted information requests ... remain available." The same package narrows scope for about 28,700 companies (6,200 of them micro and small) and adds a small mid-cap category covering another 22,500. Separately, NIS2's own review clause (Art.40) has the Commission reporting by 17 October 2027, with a legislative proposal attached "where necessary." None of it is law: the certification scheme does not exist yet, ENISA has to build a candidate "within one year as a rule," and adoption is not expected before late 2026 or 2027. If anything it sharpens the point - even the exemption being drafted is a certificate from an accredited body, not a clean scan. Do not plan around it yet.

So the practical question is not "who certifies me" but "who assembles the evidence." An accredited auditor issues the final opinion where one is required. Before that, someone has to produce the evidence base: the technical findings, the tested controls, the documented answers for everything a scan can't see. That "someone" is where a tool, a service, or a platform earns its keep. Which brings us to the tools.

Part 3: The market is full of "NIS2 checkers." Most of them can't check NIS2

Every few weeks a new product appears that promises a NIS2 assessment from a button. Strip the interface and it is almost always the same open-source stack: nmap and NSE scripts for the network, a few Kali tools, and lately an LLM agent driving all of it. There are real, shipping examples of the pattern, like mcp-kali-server ("run nmap, nxc or any other tool" from a chat client) and hexstrike-ai ("autonomously run 150+ cybersecurity tools", installable with apt install).

I like these tools. I've used all of them for years, and Pentesterra uses the same primitives under the hood. The problem is not the tools. The problem is the claim that their output is NIS2 evidence. It isn't, and here is the sourced reason, in three layers. (For the deeper teardown of why these scanners miss vulnerabilities in the first place, I wrote a separate sourced piece.)

Bar chart: of 226 NIS2 requirements, 23 are purely technical (10%), 50 are mixed (22%) and 153 are organizational (68%)
Mapped line by line against the CIR 2024/2690 annex plus six directive-level duties. Most of NIS2 is process, policy and people, not ports.

Layer 1: coverage. Most of NIS2 isn't scannable at all

First, the distinction the rest of this section rests on. A technical control is one you can prove by running something: a script, a scan, a check that returns a fact about the system - is TLS 1.2+ enforced, is this port closed, is that CVE patched. An organizational control is a matter of process, policy and people: that a SIEM exists and log sources are correctly classified, that management reviewed the risk treatment, that staff are trained, that suppliers signed incident clauses. You cannot run a check for those - the evidence is a document and an attestation, not a scan result. A mixed control is one where a scan gives a partial signal but a document still has to carry the rest. This is not a Pentesterra opinion; it is the shape of the standard.

I mapped all 226 requirement lines (the CIR annex plus directive-level duties) by nature. 153 are organizational - policies, reviews, contracts, training, physical security, governance, and everything that lives in a SIEM or a logbook rather than on a port. 50 are mixed. Only 23 are purely technical. No scanner, however clever, produces evidence that your backups were test-restored, that supplier contracts carry incident clauses, or that the board approved the security policy. The best it can do is write a sentence claiming so, which is exactly the thing an auditor exists to disprove. That caps a pure scanner at roughly 73 lines before you even ask how well it does them.

Layer 2: Nmap is a fine port scanner and a poor compliance tool

Checked against a live install: Nmap 7.98 ships 613 NSE scripts, and 211 of them are tagged intrusive. It does a genuinely good job on a handful of NIS2-relevant facts: TLS and SSH cipher strength, SMB signing, clear-text admin protocols, exposed services. That is about 8 of the 226 lines verified cleanly.

Then it runs out. There is no NSE script for SPF, DKIM, DMARC, MTA-STS, DANE, RPKI, MFA, SSO, OAuth, SAML, JWT, WebAuthn or a security.txt disclosure channel - the exact controls a compliance check needs most. Its CVE output (vulners) is banner-based: it matches the version string against a remote database. Red Hat says the consequence plainly - tools that decide "based solely on the version number ... can result in false positives as they do not account for backported security fixes." Hand that raw list to an auditor as your patch-management evidence and you have handed them a list they cannot trust. And Nmap keeps no evidence store: each run is a standalone XML file, no history, no re-test, no criticality or remediation record. CIR 6.5.2 asks for tests documented "including assessment of criticality and mitigating actions for each finding." Raw XML is not that.

Layer 3: the AI agent adds reach, and takes away evidence

An LLM driving Kali reaches further than bare Nmap. But three properties make its output unusable as evidence, no matter how good the model is.

It is not reproducible. Reproducibility is the base property of evidence. LLM inference doesn't have it, even at temperature 0. Thinking Machines Lab showed in 2025 that the same prompt sent a thousand times "returns dozens of different responses," because server-side batching changes the numerical path. An assessment you can't reproduce is a sample, not an audit.

Its results are low and inconsistent, measured. On the AutoPenBench benchmark a fully autonomous agent solved 21% of tasks (9% on real-world targets); with a human helping, 64%. A separate 400-run study pointed the same model at the same fixed target 100 times and got wildly different outcomes - one model succeeded 85 times, another 25 - with failure modes like output truncation and running out of budget. A control that "passes" on Monday and "fails" on Tuesday against an unchanged system is measuring the agent, not you.

It still runs the same tools. The agent inherits Nmap's coverage gaps and adds orchestration and prose. On the 153 organizational lines it can only generate text, and an LLM asked "is this NIS2-compliant?" will answer confidently with no basis - a hallucinated verdict wearing an assessment's clothes. Under NIS2, a false "compliant" is worse than no answer, because Art.20 makes it your management's problem.

Comparison of how many of 226 NIS2 requirements each approach can produce reproducible evidence for: Nmap+NSE 8 verified and 38 partial; AI-wrapped Kali 8 plus 58 but not reproducible; Pentesterra 46 technical evidence plus 170 documented, 10 technical gaps
Pentesterra's own numbers are honest, not perfect: 46 lines carry first-party technical evidence today, 170 more are covered by a structured questionnaire (every organizational and manual control), and the remaining 10 - all purely technical - are on the Q4 2026 roadmap. The point is completeness of the standard, not a green light from a port scan.

This is the part that isn't my opinion. NIS2 and the CIR say what evidence has to look like:

  • tests run "according to a documented test methodology" (CIR 6.5.2(b));
  • results documented "including assessment of criticality and mitigating actions for each finding" (CIR 6.5.2(c));
  • vulnerability scans "at planned intervals" with "recorded evidence" (CIR 6.10.2(b));
  • "evidence of implementation ... and the respective underlying evidence" available to the authority (Art.32(2)(g), 33(2)(f));
  • written proportionality reasoning for any "where appropriate" measure you don't apply (CIR Art.2(2)).

Methodology, reproducibility, criticality, chain of custody, documented reasoning. Those are properties of a process and a record, not of a single clever scan. A port scanner has none of them; an autonomous agent actively breaks the first two. Judge any "NIS2 tool" by that list, not by how many Kali tools it can launch.

Part 4: The full control map, all 226 lines

Below is the complete thing, nothing hidden. Every requirement of the CIR 2024/2690 annex (with lettered sub-items split where the technical answer differs) plus six directive-level duties, grouped by the 13 annex sections. For each line: its type, whether Pentesterra evidences it today, which Pentesterra capability covers it, what Nmap+NSE and an AI-Kali wrapper can do, and a short note.

This is also where the questionnaire earns its place. Every organizational and mixed line that a tool cannot reach is not left blank - it becomes a structured question, answered by the right person on your side, with the supporting document attached. That is how the assessment stays complete against the standard instead of covering only the 10% a scanner can see. The technical findings and the documented answers, with their evidence files, are what we hand to the auditor.

How to read the columns

  • Type: technical (a tool can verify it), mixed (a document plus a technical signal), org (governance: a policy, approval, review, contract or record) and manual (a recurring human task: training, testing, drills, restores, background checks). Org and manual are the two kinds of organizational control - together they are the 153 organizational lines in the chart above. A on an org or manual line is not a platform "gap": those are done off the tool and evidenced by a documented answer, so the tally below counts only unmet technical and mixed controls as gaps.
  • Pentesterra: evidenced automatically today · partial or external only · covered by the questionnaire · questionnaire, partially · not covered yet.
  • Covered by: which Pentesterra capability. Bold = wired today, a trailing ° = on the Q4 2026 roadmap. "Identity & Access" is deliberately store-agnostic: web SSO, OAuth/SAML/OIDC, JWT, WebAuthn, session and brute/lockout checks; an on-prem directory (AD/LDAP/Kerberos) where one exists; or a cloud identity provider. It never assumes you run Active Directory.
  • Nmap + NSE / AI-Kali: can verify deterministically · partial signal · can only produce an unverified claim · cannot address it.

Section D - Directive-level obligations (not in the CIR Annex)

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
Art.3(4) / 27 Register with the national authority (name, contacts, IP ranges, sectors); notify changes within 2 weeks org -
Art.20(1) Management body approves the risk-management measures, oversees them, is liable org -
Art.20(2) Management body members follow cybersecurity training manual -
Art.21(4) Take corrective measures without undue delay when non-compliance is found manual -
Art.23(4) Significant incident: early warning 24h, notification 72h, final report 1 month org -
Art.21(2)(j) Secured voice, video, text and emergency communications mixed Attack Surface° Roadmap Q4 2026. Nmap sip-methods / ssl-enum-ciphers on 5061 give a signal only
6 controls6 covered 0 evidenced · 0 partial · 6 questionnaire

Section 1 - Policy on the security of network and information systems (Art.21(2)(a))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
1.1.1 NIS security policy content (a-k): approach, objectives, resources, roles, doc retention list, topic policies list, KPIs, approval date org - (h) documentation retention list, (i) list of topic-specific policies, (j) maturity indicators are not asked
1.1.2 Policy reviewed by management at least annually and on significant change; review documented org -
1.2.1 Security responsibilities assigned to roles, communicated to management org -
1.2.2 All personnel and third parties required to apply the security policy org -
1.2.3 At least one person reports directly to management on NIS security org -
1.2.4 Dedicated roles or added duties depending on size org -
1.2.5 Segregation of conflicting duties mixed Identity & Access° Roadmap Q4 2026
1.2.6 Roles reviewed at planned intervals and on significant change org -
8 controls8 covered 0 evidenced · 0 partial · 8 questionnaire

Section 2 - Risk management policy (Art.21(2)(a))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
2.1.1 Risk-management framework, documented risk assessments, treatment plan, residual risk accepted by management org -
2.1.2(a-c) Risk methodology, risk tolerance, risk criteria org -
2.1.2(d) All-hazards risk identification incl. third parties and single points of failure mixed Attack Surface°, Attack Chain / BAS°, Compliance engine° Roadmap Q4 2026
2.1.2(e) Risk analysis using threat intelligence and known vulnerabilities mixed OSINT° Roadmap Q4 2026. Nmap only gives a vulnerability list, not a risk analysis
2.1.2(f-j) Evaluate, prioritise, monitor treatment, name owners and dates, document residual-risk acceptance org Compliance engine°
2.1.3 Treatment options consider effectiveness results, cost/benefit, asset classification, BIA org -
2.1.4 Risk assessment and treatment plan reviewed at least annually org -
2.2.1 Regular compliance review with the security policies; management informed org - Whole 2.2 subsection missing
2.2.2 Effective compliance reporting system giving management an informed view org -
2.2.3 Compliance monitoring at planned intervals and on significant change org -
2.3.1 Independent review of the approach to NIS security (people, processes, technology) org - Our row narrows it to the risk framework only
2.3.2 Reviewers have audit competence and are independent of the reviewed area org -
2.3.3 Review results reported to management; corrective action or risk acceptance org -
2.3.4 Independent reviews at planned intervals and on significant change org -
14 controls14 covered 0 evidenced · 0 partial · 14 questionnaire

Section 3 - Incident handling (Art.21(2)(b))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
3.1.1 Incident-handling policy: roles, detect, analyse, contain, recover, document, report org -
3.1.2(a-d) Incident categorisation, communication/escalation plans, roles, playbooks and contact lists org -
3.1.3 Incident roles and procedures tested and reviewed at planned intervals org - Roadmap Q4 2026
3.2.1 Procedures and tools to monitor and log activity to detect incidents org Attack Chain / BAS° Roadmap Q4 2026. No external tool can see the client's SIEM
3.2.2 Monitoring automated where feasible, minimising false positives/negatives org Attack Chain / BAS° Roadmap Q4 2026
3.2.3(a-l) Log sources: network traffic, account changes, access, auth events, privileged activity, config/backup access, security-tool logs, resources, physical access, network devices, log start/stop, environmental org Identity & Access° Roadmap Q4 2026
3.2.4 Logs reviewed; alarm thresholds; timely qualified response to alarms org Attack Chain / BAS° Roadmap Q4 2026
3.2.5 Logs retained for a predefined period, backed up and protected from tampering org - Retention asked; backup and tamper-protection of logs not asked
3.2.6 Synchronised time sources; list of logged assets; redundant logging; independent monitoring of the monitoring org - Roadmap Q4 2026
3.2.7 Logging procedures and logged-asset list reviewed org -
3.3.1 Simple mechanism for employees, suppliers, customers to report suspicious events mixed Attack Surface External half = security.txt / security contact. Nmap http-* scripts can fetch security.txt only by custom script
3.3.2 Reporting mechanism communicated to suppliers/customers; employees trained manual Attack Surface°
3.4.1 Assess suspicious events: incident or not, nature and severity org -
3.4.2(a) Predefined assessment criteria and triage org -
3.4.2(b) Quarterly assessment of recurring incidents manual -
3.4.2(c-d) Review logs for assessment; log correlation and analysis process org - Roadmap Q4 2026
3.4.2(e) Reassess and reclassify events on new information org -
3.5.1-3.5.2 Respond per documented procedures: containment, eradication, recovery org -
3.5.3 Communication plans with CSIRT / authority and internal/external stakeholders org -
3.5.4 Log incident-response activities and record evidence org -
3.5.5 Test incident-response procedures at planned intervals org - Roadmap Q4 2026
3.6.1 Post-incident review with root cause and lessons learned manual Identity & Access°
3.6.2 Post-incident reviews feed improvements to measures and procedures org -
3.6.3 Check at planned intervals that incidents led to post-incident reviews manual -
24 controls24 covered 1 evidenced · 0 partial · 23 questionnaire

Section 4 - Business continuity and crisis management (Art.21(2)(c))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
4.1.1 Business continuity and disaster recovery plan org -
4.1.2(a-h) Plan content: scope, roles, contacts, activation, recovery order and objectives, resources, return from temporary measures org -
4.1.3 Business impact analysis and resulting continuity requirements org -
4.1.4 BC and DR plans tested, reviewed, lessons incorporated manual Identity & Access°
4.2.1 Backup copies and sufficient redundant resources org -
4.2.2(a) Backup plan: recovery times org -
4.2.2(b) Backups complete and accurate incl. configuration and cloud data org Attack Surface°
4.2.2(c) Backups in a safe location, not in the same network, at sufficient distance org Attack Surface° Roadmap Q4 2026
4.2.2(d) Physical and logical access control to backup copies org - Roadmap Q4 2026
4.2.2(e) Restoring data from backups manual Attack Surface°
4.2.2(f) Backup retention periods org -
4.2.3 Regular backup integrity checks manual -
4.2.4(a-d) At least partial redundancy of systems, facilities, personnel, communication channels org - Roadmap Q4 2026
4.2.5 Resource monitoring informed by backup and redundancy requirements org -
4.2.6 Regular recovery testing of backups and redundancies, documented manual Attack Surface°
4.3.1-4.3.2 Crisis-management process: roles, communication with authorities, security during crisis org -
4.3.3 Process to use information from CSIRTs / authorities on threats and vulnerabilities org - Roadmap Q4 2026
4.3.4 Crisis plan tested and updated manual -
18 controls18 covered 0 evidenced · 0 partial · 18 questionnaire

Section 5 - Supply chain security (Art.21(2)(d))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
5.1.1 Supply-chain security policy; entity's role in the chain communicated org DevGuard°
5.1.2(a-d) Supplier selection criteria: security practices, ability to meet specs, product quality, diversification mixed Attack Surface° Roadmap Q4 2026
5.1.3 Consider EU coordinated supply-chain risk assessments org DevGuard°
5.1.4(a,d,e,f,g) Contracts: security requirements, incident notification, audit rights, vulnerability handling, subcontracting org -
5.1.4(b,c,h) Contracts: supplier staff training/certification, background checks, obligations on termination manual -
5.1.5 Criteria applied in supplier selection and procurement org Attack Surface°
5.1.6 Monitor and act on changes in suppliers' security practices mixed Attack Surface Supplier screening re-scan. Nmap/Kali need authorisation to scan a third party
5.1.7(a) Monitor SLA implementation reports org -
5.1.7(b) Review incidents related to suppliers' products and services org - Roadmap Q4 2026. Capability exists (breach/ransomware monitoring), not wired to NIS2
5.1.7(c-d) Unscheduled reviews; analyse risk of supplier changes mixed Attack Surface° Roadmap Q4 2026
5.2 Directory of direct suppliers with contacts and supplied ICT products/services org Attack Surface°
11 controls11 covered 1 evidenced · 0 partial · 10 questionnaire

Section 6 - Security in acquisition, development and maintenance (Art.21(2)(e))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
6.1.1 Manage risks from acquiring critical ICT products/services across their life cycle org -
6.1.2(a) Security requirements for acquired ICT org -
6.1.2(b) Security updates for the whole lifetime, or replacement at end of support mixed Attack Surface° External EOL detection is a proxy; the contractual half is not asked
6.1.2(c) Information on hardware and software components (SBOM) org DevGuard Roadmap Q4 2026
6.1.2(d) Documentation of security functions and secure configuration org -
6.1.2(e-f) Assurance and validation that delivered ICT meets requirements; documented mixed Web/API PT°, Compliance engine° Roadmap Q4 2026
6.1.3 Acquisition processes reviewed org -
6.2.1 Secure development rules for all phases, in-house and outsourced mixed DevGuard°
6.2.2(a) Security requirements analysis at specification/design org -
6.2.2(b) Secure engineering and secure coding principles technical DevGuard DevGuard SAST. Kali has no SAST by default (semgrep can be added)
6.2.2(c) Security requirements for development environments mixed DevGuard°
6.2.2(d) Security testing processes in the development life cycle technical Web/API PT
6.2.2(e-f) Select, protect, sanitise and anonymise test data org DevGuard Roadmap Q4 2026
6.2.3 Outsourced development follows supply-chain and acquisition policies org -
6.2.4 Secure development rules reviewed org -
6.3.1 Establish, document, implement and monitor security configurations technical Attack Surface, Web/API PT° External only: headers, TLS, cookies
6.3.2(a) Secure configurations defined for hardware, software, services, networks mixed Attack Surface, Identity & Access°, VM°
6.3.2(b) Processes and tools enforce the secure configuration over the lifetime mixed DevGuard° Roadmap Q4 2026
6.3.3 Configurations reviewed at planned intervals org -
6.4.1-6.4.2 Change management for releases, modifications, emergency changes; documented, tested, impact-assessed manual Compliance engine°
6.4.3 Emergency changes documented with the reason procedures were skipped org -
6.4.4 Change procedures reviewed org -
6.5.1 Security-testing policy and procedures org -
6.5.2(a) Need, scope, frequency and type of tests set on a risk basis org Attack Chain / BAS°
6.5.2(b) Tests follow a documented methodology covering relevant components technical VM, Web/API PT Nmap = one test type. AI agent runs are not reproducible
6.5.2(c) Document type, scope, time, results, criticality and mitigation per finding technical VM, Web/API PT Nmap gives raw XML, no criticality or mitigation. LLM summaries are unverified
6.5.2(d) Mitigating actions applied for critical findings org - Roadmap Q4 2026. Missing as a distinct check
6.5.3 Security-testing policy reviewed org -
6.6.1(a) Security patches applied within reasonable time technical Attack Surface, VM Nmap vulners = banner-version matching, high false-positive rate
6.6.1(b) Patches tested before production manual -
6.6.1(c) Patches from trusted sources, integrity-checked org - Roadmap Q4 2026
6.6.1(d), 6.6.2 Compensating measures and documented justification when a patch is not applied org -
6.7.1 Protect network and information systems from cyber threats mixed Attack Surface°, VM° Roadmap Q4 2026
6.7.2(a) Network architecture documented and up to date org -
6.7.2(b) Controls protecting internal network domains from unauthorised access mixed Identity & Access° Roadmap Q4 2026. Kali has responder / netexec; Nmap has smb-* and broadcast scripts
6.7.2(c) Block access and communication not required for operations technical Attack Surface External only
6.7.2(d) Controls for remote access incl. service providers technical Attack Surface
6.7.2(e) Security-administration systems not used for other purposes org -
6.7.2(f) Forbid or disable unneeded connections and services technical Attack Surface
6.7.2(g) Only authorised devices may access the network technical Attack Chain / BAS, VM, Web/API PT Roadmap Q4 2026
6.7.2(h) Service-provider connections only after authorisation and time-limited org -
6.7.2(i) Trusted, isolated, authenticated, integrity-protected channels between systems technical Attack Surface, Identity & Access°, VM° Wired only for external/web TLS
6.7.2(j) Implementation plan for transition to latest-generation network protocols (IPv6) org - Roadmap Q4 2026
6.7.2(k) Plan to deploy modern e-mail security standards (SPF, DKIM, DMARC, STARTTLS, DANE, MTA-STS) mixed Attack Surface Roadmap Q4 2026. Top quick win. Nmap has smtp-open-relay + STARTTLS ciphers only, no SPF/DMARC
6.7.2(l) DNS security best practice and Internet routing hygiene (DNSSEC, RPKI) mixed Attack Surface Roadmap Q4 2026. Nmap: dns-recursion, dns-zone-transfer, dns-nsec-enum; no RPKI
6.7.3 Network-security measures reviewed org -
6.8.1 Segment systems into zones per risk; segment from third-party networks mixed Attack Surface, Attack Chain / BAS° Nmap can do it only if someone runs it from inside each zone
6.8.2(a-b) Consider system relationships; grant zone access on security assessment org Attack Chain / BAS°
6.8.2(c) Critical / safety systems kept in secured zones mixed Attack Chain / BAS° Roadmap Q4 2026
6.8.2(d) Deploy a DMZ mixed Attack Surface
6.8.2(e) Restrict communication between and within zones to what is necessary mixed Attack Chain / BAS° Roadmap Q4 2026
6.8.2(f-g) Separate administration network; segregate admin channels technical Attack Surface External half only
6.8.2(h) Separate production from development and test, incl. backups mixed Attack Surface°, Web/API PT° Roadmap Q4 2026
6.8.3 Segmentation reviewed org -
6.9.1-6.9.2 Protection against malicious/unauthorised software; EDR kept updated mixed Attack Chain / BAS° Roadmap Q4 2026
6.10.1 Obtain vulnerability information, evaluate exposure, manage vulnerabilities technical Attack Surface
6.10.2(a) Monitor vulnerability information via CSIRTs, authorities, suppliers org - Roadmap Q4 2026
6.10.2(b) Vulnerability scans at planned intervals with recorded evidence technical VM, Web/API PT Nmap/agent: no evidence store or history
6.10.2(c) Address critical vulnerabilities without undue delay technical VM, Compliance engine°
6.10.2(d) Vulnerability handling consistent with change, patch, risk, incident management org -
6.10.2(e) Vulnerability-disclosure procedure per national CVD policy mixed Attack Surface
6.10.3 Mitigation plan for impactful vulnerabilities, else documented reason mixed Compliance engine° Roadmap Q4 2026
6.10.4 Vulnerability-information channels reviewed org -
63 controls62 covered1 technical gap 12 evidenced · 18 partial · 32 questionnaire · 1 technical/mixed gap

Section 7 - Effectiveness assessment (Art.21(2)(f))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
7.1 Policy and procedures to assess effectiveness of measures org -
7.2(a-f) What is measured, methods, when, who measures, when and who analyses org Phishing°, Compliance engine°
7.3 Effectiveness policy reviewed org -
3 controls3 covered 0 evidenced · 0 partial · 3 questionnaire

Section 8 - Cyber hygiene and security training (Art.21(2)(g))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
8.1.1 Employees, management, suppliers aware of risks and apply cyber hygiene org Attack Surface° Roadmap Q4 2026
8.1.2(a-c) Awareness programme: repeated, covers new joiners, aligned to policy, covers threats and hygiene org Phishing°
8.1.3 Awareness programme effectiveness tested technical Phishing Connector is presence-only: never returns a gap
8.2.1-8.2.2 Identify roles needing security skills; role-based training programme manual Attack Chain / BAS°
8.2.3(a-c) Training content (secure config, threats, behaviour) and effectiveness assessed manual -
8.2.4 Training when staff move into security-relevant roles manual -
8.2.5 Training programme updated and run periodically manual -
7 controls7 covered 0 evidenced · 1 partial · 6 questionnaire

Section 9 - Cryptography (Art.21(2)(h))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
9.1 Cryptography policy aligned to asset classification and risk org -
9.2(a) transit Required cryptographic strength for data in transit technical Attack Surface, VM°
9.2(a) rest Required cryptographic strength for data at rest org -
9.2(b) Approved protocols, algorithms, cipher strength; crypto agility technical Attack Surface, DevGuard° Nmap ssl-enum-ciphers / ssh2-enum-algos are genuinely good here
9.2(c)(i-xii) Key management life cycle incl. certificates, rotation, revocation, compromise mixed DevGuard°, Attack Surface° Roadmap Q4 2026
9.3 Crypto policy reviewed against state of the art org -
6 controls6 covered 1 evidenced · 1 partial · 4 questionnaire

Section 10 - Human resources security (Art.21(2)(i))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
10.1.1 Staff and suppliers understand and commit to security responsibilities org Attack Surface°
10.1.2(a) Mechanisms so everyone follows cyber-hygiene practice mixed Phishing° Roadmap Q4 2026
10.1.2(b) Privileged users aware of their responsibilities org Identity & Access°
10.1.2(c) Management understands its role org -
10.1.2(d) Hiring qualified personnel (references, vetting, certification checks, tests) manual -
10.1.3 Role assignment reviewed at least annually org -
10.2.1-10.2.3 Background verification per role criteria, before role starts; policy reviewed org -
10.3.1-10.3.2 Security duties that survive termination defined in contracts org -
10.4.1-10.4.2 Disciplinary process for policy violations; reviewed org -
9 controls9 covered 0 evidenced · 0 partial · 9 questionnaire

Section 11 - Access control (Art.21(2)(i),(j))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
11.1.1-11.1.3 Logical and physical access-control policy (persons, systems, authenticated access); reviewed org -
11.2.1 Provide, modify, remove and document access rights per policy org Identity & Access°
11.2.2(a) Need-to-know, least privilege, separation of duties mixed Identity & Access° Roadmap Q4 2026
11.2.2(b) Access rights changed on termination or role change mixed Identity & Access, OSINT, Attack Surface°
11.2.2(c) Access authorised by the relevant persons org -
11.2.2(d) Third-party access limited in scope and duration org -
11.2.2(e) Register of access rights granted org -
11.2.2(f) Logging of access-rights management org -
11.2.3 Periodic access-rights review, documented org -
11.3.1 Policy for privileged and system-administration accounts org Identity & Access°
11.3.2(a) Strong identification, authentication (MFA) and authorisation for privileged accounts mixed Identity & Access° External wiring only checks admin services are not exposed
11.3.2(b) Dedicated accounts used only for administration mixed Identity & Access° Roadmap Q4 2026
11.3.2(c) Individualise and restrict admin privileges mixed Identity & Access° Roadmap Q4 2026
11.3.2(d) Admin accounts only connect to administration systems mixed Identity & Access° Roadmap Q4 2026
11.3.3 Privileged access rights reviewed, documented org Identity & Access°
11.4.1 Restrict and control use of system-administration systems mixed Attack Surface
11.4.2(a-b) Admin systems used only for admin; logically separated mixed Identity & Access° Roadmap Q4 2026
11.4.2(c) Admin systems protected by authentication and encryption technical VM, Identity & Access°
11.5.1 Manage full identity life cycle org Identity & Access°
11.5.2(a-b) Unique identities, each linked to one person mixed Web/API PT, Identity & Access°
11.5.2(c) Oversight of system (non-human) identities mixed Identity & Access°, DevGuard° Roadmap Q4 2026
11.5.2(d) Logging of identity management org - Roadmap Q4 2026
11.5.3 Shared identities only when needed, approved and documented mixed Identity & Access° Roadmap Q4 2026
11.5.4 Review identities and deactivate unused ones without delay technical Identity & Access° Roadmap Q4 2026
11.6.1 Secure authentication procedures and technologies mixed Identity & Access
11.6.2(a) Authentication strength fits the asset classification mixed Identity & Access° Roadmap Q4 2026
11.6.2(b) Secret authentication information handled confidentially mixed DevGuard, Identity & Access° DevGuard no_hardcoded_credentials / no_hardcoded_secrets
11.6.2(c) Credentials changed initially, at intervals, on suspected compromise technical Identity & Access°, OSINT° Roadmap Q4 2026. Nmap http-default-accounts; Kali hydra (risky on production)
11.6.2(d) Reset and lockout after a number of failed log-in attempts technical Identity & Access° Roadmap Q4 2026. Brute-force testing can lock real users; must be rate-limited
11.6.2(e) Terminate sessions after inactivity technical Web/API PT, Identity & Access° Current wiring checks cookie flags, not the timeout itself
11.6.2(f) Separate credentials for privileged / admin accounts mixed Identity & Access° Roadmap Q4 2026
11.6.3 State-of-the-art authentication methods mixed Identity & Access° Roadmap Q4 2026
11.6.4 Authentication procedures reviewed org -
11.7.1 MFA or continuous authentication where appropriate mixed Attack Surface, Identity & Access° External wiring = VPN appliance exposure only; cannot see MFA
11.7.2 Authentication strength appropriate to asset classification mixed Identity & Access° Roadmap Q4 2026
35 controls26 covered9 technical gaps 3 evidenced · 6 partial · 17 questionnaire · 9 technical/mixed gaps

Section 12 - Asset management (Art.21(2)(i))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
12.1.1-12.1.3 Asset classification levels, every asset classified, aligned with BC objectives, reviewed org -
12.2.1-12.2.2(a-b) Asset-handling policy over the life cycle incl. deletion and destruction org -
12.2.2(c) Secure transfer of assets org -
12.2.3 Handling policy reviewed org -
12.3.1 Removable-media policy communicated org -
12.3.2(a-b) Technical block of removable media; autorun disabled; media scanned mixed -
12.3.2(c-d) Protect portable media in transit; encryption org -
12.3.3 Removable-media policy reviewed org -
12.4.1 Complete, accurate, current inventory with traceable changes mixed Attack Surface
12.4.2(a-b) Inventory lists operations/services and supporting systems mixed Attack Surface° Roadmap Q4 2026
12.4.3 Inventory reviewed with change history mixed Compliance engine° Roadmap Q4 2026
12.5 Assets returned / deleted on termination, or blocked from access org Identity & Access
12 controls12 covered 0 evidenced · 2 partial · 10 questionnaire

Section 13 - Environmental and physical security (Art.21(2)(c),(e),(i))

CIR ref Control Type Pentesterra Covered by Nmap + NSE AI-Kali Note
13.1.1-13.1.2(a-b) Protect against utility failures; utility redundancy org -
13.1.2(c) Protect power and telecom cabling from interception and damage org -
13.1.2(d) Monitor utilities, report threshold events org -
13.1.2(e) Emergency-supply contracts (e.g. generator fuel) org -
13.1.2(f) Monitor, maintain, test supply of power, cooling, connectivity manual -
13.1.3 Utility protections tested and reviewed manual -
13.2.1-13.2.2 Protection against physical/environmental threats; thresholds; monitoring org Attack Surface°
13.2.3 Environmental protections tested and reviewed manual -
13.3.1-13.3.2 Security perimeters, entry controls, physical security, continuous monitoring org Attack Surface°
13.3.3 Physical access controls tested and reviewed manual -
10 controls10 covered 0 evidenced · 0 partial · 10 questionnaire

Reach is not evidence

Here is the trap in every "free NIS2 scanner" pitch: it counts what a tool touches, not what it can prove. Run against all 226 lines and the gap is not subtle. Nmap reaches 46 controls, an AI-Kali agent 66, and neither reaches more than a handful of the 153 organizational controls that actually decide NIS2. But reaching a control is not evidencing it. Nmap's output is deterministic, so it verifies 8; the AI agent's runs are not reproducible, so however many tools it fires, it proves none. Pentesterra evidences 216 - the 46 technical with first-party proof and the 170 organizational and manual through the questionnaire, each with the underlying evidence attached to the assessment. That is what you are paying for: not a bigger scan than Nmap, but an audit-ready evidence base across the whole standard, including the 68% no scanner can reach.

Reach versus admissible evidence, out of 226 NIS2 controls: Pentesterra reaches 216 and evidences all 216; Nmap+NSE reaches 46 and evidences 8; an AI-wrapped Kali reaches 66 but evidences 0, because its output is not reproducible
The AI agent reaches the most and proves the least: 66 controls touched, admissible evidence for none, because its runs are not reproducible. Nmap reaches 46 and verifies 8. Pentesterra evidences 216 - the technical band with first-party proof, the organizational majority through the questionnaire, each with the underlying evidence attached. An auditor accepts evidence, not reach.

In closing: what Pentesterra actually does here

I'll be straight about our own numbers, because Part 3 was about honesty. Today we evidence 46 of those 226 lines with first-party technical proof and cover the organizational and manual controls - 170 lines - through the structured questionnaire, so the assessment is complete against the whole standard. Only 10 lines, all purely technical, are not yet automated; those are on the Q4 2026 roadmap and marked as such in the table above. We do not issue a compliance verdict and we are not your auditor. What we do is produce the evidence base an auditor needs, and an indicative readiness view for your own management.

Concretely, Pentesterra brings to a NIS2 assessment:

  • First-party technical evidence on the technical and mixed controls - from external exposure, network and web/API testing, DevGuard code and supply-chain scanning, identity and access testing, and phishing simulation. Real findings with proof, not a version guess. And we attach that evidence - the findings, the proof-of-concept, the scan records - to the assessment, so what reaches the auditor is the underlying evidence the law asks for, not a summary.
  • A structured questionnaire for the organizational controls, routed to the right people, with document attachments, so the assessment is complete against the whole standard and not just the scannable slice. This is the part that keeps you from quietly failing on the 68% no scanner can see.
  • One matrix that shows, per control, what the evidence says, what the answer says, and where the gaps are - plus an indicative readiness score and a white-label gap report you can hand onward. Every gap becomes a concrete next step rather than a vague "improve security."

And you can use it two ways. As a service, where we run the assessment and deliver the evidence base and gap report. Or as a platform, where your own team, or your partner's, runs the self-assessment, launches the technical modules, and manages the evidence in one place. Same engine either way.

NIS2 is not going back in the drawer, and the tools promising to make it disappear mostly can't produce what a regulator will ask for. If you want to see the assessment on your own scope, the compliance module is here, and I'm happy to walk a real environment through it.


Sources

  • Directive (EU) 2022/2555 (NIS2), full text: eur-lex.europa.eu - scope (Art.2, 3), governance and liability (Art.20), measures (Art.21), incident reporting (Art.23), certification (Art.24), registration (Art.27), supervision and fines (Art.32-34)
  • Commission Implementing Regulation (EU) 2024/2690, full annex: eur-lex.europa.eu
  • Commission refers IE, ES, FR, NL to the Court of Justice (Jul 2026): ec.europa.eu, IP/26/1499
  • Commission cybersecurity package / NIS2 amendment Q&A (Jan 2026) - cyber-posture certificate, "presumption of conformity", ~28,700 entities eased, 22,500 small mid-caps: digital-strategy.ec.europa.eu
  • Analysis of the Jan 2026 NIS2 amendments - targeted-audit exemption for certificate holders, "on-site inspections, security scans ... remain available": Covington, Inside Privacy
  • Certification only where done by a third-party conformity assessment body (Art.78(2) revised Cybersecurity Act): Reed Smith
  • NIS2 Art.40 review clause - Commission report by 17 October 2027, legislative proposal "where necessary": nis2resources.eu
  • Belgium CCB, 18 April 2026 conformity deadline: ccb.belgium.be
  • Germany BSIG proof duty (§39): gesetze-im-internet.de
  • Italy ACN, NIS second phase: acn.gov.it
  • Red Hat, security backporting and version-based false positives: access.redhat.com
  • Thinking Machines Lab, Defeating Nondeterminism in LLM Inference (2025): thinkingmachines.ai
  • Gioacchini et al., AutoPenBench (arXiv 2410.03225): arxiv.org
  • Erdem, A 400-Run Empirical Study of LLM Penetration Testing Consistency (arXiv 2605.30096, 2026): arxiv.org
  • The "AI + Kali" product category: mcp-kali-server, hexstrike-ai
  • Deeper teardown of scanner accuracy: Why Nmap and Open-Source Scanners Don't Detect Vulnerabilities Reliably
Share on LinkedInhttps://pentesterra.com/blog/nis2-who-must-comply-and-what-actually-proves-it

Take Control of Your Attack Surface.

Talk to us about your environment - network, web, cloud, or on-prem.