Italy's NIS2 baseline deadline has passed: a pre-audit checklist for software vendors on AWS, Azure or Google Cloud
ACN's baseline security measures under Legislative Decree 138/2024 were due in October 2026 for entities listed in 2025, and ACN can now verify them. Here is what a software vendor running on AWS, Azure or Google Cloud should check, measure by measure, before an ACN inspection or a customer audit.
The deadline is 18 months from the day you received ACN's inclusion notice, which ACN itself dates to October 2026 for entities listed in 2025. Italian practitioners use 31 October 2026 as the reference, but your own date comes from the PEC you received. Entities first listed in 2026 have until 31 July 2027 (ACN FAQ MSB.3).
The rulebook is ACN Determinazione 379907 of 19 December 2025. It replaced Determinazione 164179 of April 2025 and applies from 15 January 2026. Annex 1 holds 37 baseline measures for important entities, Annex 2 holds 43 for essential entities (ACN).
ACN can now verify, and the rules differ by category. Under Article 36 of Legislative Decree 138/2024, essential entities can face on-site, remote and random inspections at any time. Important entities face them only when ACN has evidence or indications of a violation (Gazzetta Ufficiale).
A hyperscaler does not carry your measures for you. AWS, Azure and Google Cloud cover the physical layer. Accounts, MFA, logs, backups, supplier inventory and incident plans sit on your side of the shared responsibility line, and that is where an inspector will look.
The checklist below maps the ACN measures that touch a cloud-hosted product to the evidence you should be able to export today. Qovery helps with the deployment layer (it runs in your own cloud account, with RBAC and an audit log), but it does not make anyone NIS2-compliant.
Every Italian CTO I spoke to in October asked me some version of the same question: "the deadline has passed, what will ACN actually ask for?" The short answer is documents. Inventories, plans, policies, procedures and registers, plus proof that the technical controls behind them are switched on.
Italy is ahead of most of Europe here. Its NIS2 law, Legislative Decree 4 September 2024, n. 138, has been in force since 16 October 2024, and ACN has already published a detailed, numbered set of baseline measures. Numbered measures make a measure-by-measure pre-audit checklist possible. Most member states still only have the directive's ten headings.
I wrote this for software vendors whose product runs on AWS, Azure or Google Cloud and who are either a NIS entity themselves or sell to one. If you are looking for the separate regime that decides which clouds an Italian public administration may buy, that is ACN cloud qualification, and we cover it in our guide to ACN-qualified cloud platforms. People often confuse the two, but they answer different questions.
Has Italy's NIS2 deadline really passed, and what does "inspection phase" mean?
Yes, for entities that ACN listed in 2025. ACN's Determinazione 379907/2025 sets the baseline security measures deadline at 18 months from receipt of the inclusion notice, and ACN's own NIS page dates that to October 2026 (ACN). The incident notification obligation started earlier, 9 months after inclusion, so January 2026.
Here is the timeline as the primary sources state it.
A note on "31 October". ACN's texts say "18 months from receipt" and "October 2026". The 31 October date is the one Italian consultancies and trade press use (PuntoSicuro). Check the date on your own PEC before you argue about a few days with an inspector.
The phrase "inspection phase" refers to legal powers that now apply. I did not find a dated ACN announcement or an inspection calendar. What exists is the legal power, in Articles 35 to 38 of the decree, and ACN's August 2026 FAQs explaining how it will use it: monitoring, analysis and support; verifications and inspections; enforcement measures; and administrative sanctions, applied "gradually" and based on risk (ACN FAQ MVE). Once the deadline passed, nothing stops ACN from using all four.
What can ACN do during an inspection?
ACN can check the documents you filed, inspect you on site or remotely (including random checks), and require access to data, documents and other information. Article 36 of Legislative Decree 138/2024 lists these three powers. Article 36(2) limits them for important entities to cases where ACN "acquires or receives evidence, indications or information" of a possible violation (Gazzetta Ufficiale).
Article 37 goes further. ACN can ask you to provide "the data that demonstrate the implementation of cybersecurity policies, such as the results of security audits and the related evidence", and it can order you to run periodic or targeted security audits, in particular after a significant incident.
The sanctions sit in Article 38:
Entity type
Maximum fine
Minimum fine
Essential entity (private)
€10,000,000 or 2% of worldwide annual turnover, whichever is higher
One twentieth of the maximum
Important entity (private)
€7,000,000 or 1.4% of worldwide annual turnover, whichever is higher
Article 23 adds personal accountability. Management and governing bodies must approve how the risk management measures are implemented, oversee them, follow cybersecurity training, and they "are responsible for violations" of the decree. In a vendor with a small board, that is your CEO and CTO by name.
Does running on AWS, Azure or Google Cloud cover any of the ACN measures for you?
It covers the physical and infrastructure layer of the provider's own services, and very little of what ACN's baseline asks you to document. Every hyperscaler draws the same shared responsibility line: the provider secures the data centres, hardware, hypervisor and managed service internals, and you secure identities, configuration, data, logs and the way you change production (AWS, Microsoft, Google Cloud).
Two Italy-specific points change how you treat the hyperscaler:
The cloud provider is a supplier you must inventory and assess. Measure ID.AM-04 requires an up-to-date inventory of IT services delivered by suppliers, "including cloud services". GV.SC-04 requires a supplier inventory with a contact person and supply type. GV.SC-07 requires a documented risk assessment per supply (ACN Annex 1).
Incident notification depends on the cloud model. When an incident at a cloud provider that is itself a NIS entity hits a client NIS entity, both must notify CSIRT Italia, "except where the cloud service is IaaS or hosting of the client's infrastructure", in which case only the client notifies (ACN FAQ ISB.F.3). If you run your product on EC2, EKS, AKS or GKE, the notification is yours.
Are you a NIS entity, a supplier to one, or both?
Check this first, because it decides which checklist applies. A software vendor usually sits in one of three positions:
You received ACN's inclusion notice. You are an essential or important entity and must apply Annex 1 or Annex 2 of Determinazione 379907 on your own systems. If you registered in a digital infrastructure or ICT service management category, also read Commission Implementing Regulation (EU) 2024/2690, which sets technical requirements directly for cloud computing service providers, managed service providers and similar digital providers.
You sell software to NIS entities. You are not inspected by ACN for your own systems, but your customers must put security requirements into their contracts with you (GV.SC-05), verify your compliance periodically (GV.SC-07 point 2), and require you to report security events that affect them promptly (ACN FAQ ISB.F.3). Expect a questionnaire that copies ACN's measure codes.
Both. Common for Italian SaaS companies serving utilities, health or transport.
The rest of this checklist is written for the first case. If you are in the second, use it to prepare the answers your customer's procurement team will ask for.
Which ACN baseline measures matter most for a cloud-hosted product?
For a product running on a hyperscaler, roughly fifteen of the 37 important-entity measures carry most of the technical weight: identities, MFA, privileged access, logging, backups, patching, secure development, supplier management and incident response. The table maps each one to what usually goes wrong in a cloud account and the evidence I would have ready. Measure codes and wording come from Annex 1 of Determinazione 379907.
ACN measure
What it requires (paraphrased)
Typical cloud gap
Evidence to have ready
GV.SC-04
Up-to-date inventory of suppliers with potential security impact, with contact and supply type
CI/CD, observability and deployment SaaS missing from the list
Supplier register with owner, contact, supply type, last review date
GV.SC-05
Security requirements written into tenders, contracts and agreements
Click-through terms with no security clauses
Signed DPAs, security addenda, breach notification clauses per supplier
GV.SC-07
Documented risk per supply: supplier access level, data access, outage impact, recovery time and cost, supplier roles
No record of which vendors hold production credentials
Per-supplier risk sheet, including what IAM role each vendor assumes in your account
ID.AM-02
Inventory of services, systems and applications, including those reached via API
Clusters and accounts created outside Terraform
Exported inventory of accounts, clusters, environments and services
ID.AM-04
Inventory of supplier IT services, cloud included
Managed services used but not listed (queues, CDN, email)
Documented remote activities, list of systems reachable remotely, maintained perimeter systems
Bastions and kubectl access nobody listed
Remote access register, firewall and security group reviews
DE.CM-01
Detection tools for significant incidents, documented expected service levels (SL)
No written SLs, so incident IS-3 cannot be detected
SL document per service, alert rules mapped to it
RS.MA-01
Documented incident management and CSIRT notification plan
Plan does not cover the 24h and 72h steps
Incident plan with notification roles, tabletop record
Essential entities get six more measures in Annex 2, including configuration management (PR.PS-01), resilience mechanisms (PR.IR-03), network and data flow representations (ID.AM-03) and recovery communications (RC.CO-03). Their PR.DS-11 also adds encryption or physical protection of backups and periodic restore tests (ACN Annex 2).
A word on DE.CM-01. Annex 3 defines the baseline significant incidents for important entities in three lines: loss of confidentiality of data to the outside (IS-1), loss of integrity with external impact (IS-2), and breach of the expected service levels you set under DE.CM-01 (IS-3) (ACN Annex 3). If you never wrote your service levels down, you cannot tell ACN when IS-3 happened. I would fix this one first. It takes an afternoon.
Get your deployment evidence ready before ACN asks for it.
Qovery deploys your applications into your own AWS, GCP, Azure or Scaleway account, or your existing Kubernetes cluster, with per-environment RBAC and an audit log of every change. Talk to us about mapping it to your NIS2 evidence pack.
What does the incident clock look like when your product runs in someone else's data centre?
The clock is the same as in the directive. Article 25 of Legislative Decree 138/2024 requires a pre-notification to CSIRT Italia within 24 hours of becoming aware of a significant incident, a notification within 72 hours with an initial assessment of severity and impact, intermediate reports on request, and a final report within one month of the notification (Gazzetta Ufficiale).
What changes on a hyperscaler is where the facts live. To file a credible 72-hour notification you need to know:
what changed in production in the hours before the incident, and who changed it
which accounts, clusters and environments are affected
whether the cause sits with you or with the provider (check the provider's health dashboard and your support case history)
whether personal data is involved, because a GDPR breach notification to the Garante can run in parallel on its own 72-hour clock
In most vendors I talk to, those answers are spread across the cloud audit log, the CI system, a deployment tool, Kubernetes events and Slack. The rehearsal ACN's RS.MA-01 implies is to rebuild one past incident from those sources and time how long it takes. If it takes more than a few hours, your 24-hour pre-notification will be a guess.
What is the pre-audit checklist, step by step?
Work through it in this order. Each step produces a document an inspector or a customer auditor can read.
Confirm your status and date. Find the ACN inclusion PEC, note whether you are essential or important, and compute your 18-month date. Record the category you registered under.
Get the management body sign-off. Article 23 requires the board to approve the implementation of the measures and to follow training. Minute it, with the date and the list of attendees trained.
Name the security organisation. GV.RR-02 in the annexes requires a defined security organisation, including the CSIRT contact person and substitutes. Put names in a document, not a slide.
Build the three inventories. Systems and services (ID.AM-02), supplier services including cloud (ID.AM-04), and suppliers with contacts (GV.SC-04). Generate them from your accounts and Terraform state where you can, so they stay current.
Assess each supplier. For your cloud provider, CI, registry, observability stack and deployment platform, write down the access level, data access, outage impact and recovery time (GV.SC-07). The question "can this vendor deploy to production without us?" belongs in every sheet.
Fix identities. Individual accounts only, MFA on everything that reaches production, separate admin identities, a dated access review (PR.AA-01, -03, -05).
Decide log retention and centralise. Enable the audit logs of every account and cluster, log privileged and remote access, ship them to one place, and document the retention period chosen after your risk assessment (PR.PS-04).
Prove backups. Backups of data and configuration, an offline or isolated copy for relevant systems, and a restore test report with a date and duration (PR.DS-11).
Patch to supported versions. List runtime, OS, database and Kubernetes versions with their end-of-support dates (PR.PS-02). Upstream Kubernetes supports each minor release for about 14 months (Kubernetes), so a cluster you skipped upgrading last year is likely outside it.
Write the service levels and the incident plan. DE.CM-01 service levels, RS.MA-01 incident plan with the 24h, 72h and one-month steps, and a tabletop exercise record.
Collect the policies. ACN's FAQ lists the document types it expects: lists, inventories, plans, policies, procedures and registers (ACN FAQ). Put them in one folder with owners and review dates.
Dry-run an Article 37 request. Ask someone outside the platform team to request "the evidence that PR.AA-03 and PR.PS-04 are implemented" and time how long the answer takes.
Where do deployment pipelines usually fail an audit?
The deployment path is where I see the most missing evidence, because it touches four measures at once: PR.AA-05 (who can change production), PR.PS-04 (is that access logged), GV.SC-07 (which vendor holds the keys) and DE.CM-01 (can you link a change to an incident).
The common failure patterns:
A long-lived cloud access key with admin rights stored in the CI system, used by every pipeline.
A deployment SaaS that runs your workloads in its own cloud account, so the data plane, its logs and its region are outside your accounts and outside your direct evidence.
Engineers with kubectl against production through a shared kubeconfig, so the cluster audit log shows one identity for everyone.
Preview environments seeded with production data, which turns a test system into a "relevant system" without anyone noticing.
Coding agents given an engineer's own credentials, so their actions are logged under that engineer's name.
Here is how the common deployment models compare against the ACN evidence questions.
BYOC platform in your account (Qovery, Northflank, Porter)
Your account or your cluster
Scoped roles you grant
Platform audit log plus your cloud audit log
Medium: control plane only
Manual console and kubectl
Your account
Individual engineers
Cloud audit logs only, no intent
Low supplier risk, high internal risk
Heroku, Render and Vercel are good products, and plenty of NIS entities use them for non-relevant systems. For systems in scope, check whether you can produce ACN's evidence without opening a ticket with someone else.
Where does Qovery fit in an Italian NIS2 evidence pack?
Qovery is an Agentic DevOps platform that deploys your applications on Kubernetes inside your own AWS, GCP, Azure or Scaleway account, or on an existing cluster, so workloads, data, logs and the cloud bill stay in your accounts. The Qovery control plane is separate from that data plane. For ACN purposes, Qovery is a supplier you add to your GV.SC-04 inventory, with access limited to what you grant in your cloud account.
On the deployment layer it gives you evidence for several of the measures above:
PR.AA-05: per-environment RBAC, so developers can deploy to staging without holding production rights (Qovery RBAC).
PR.PS-04: an audit log of who changed which environment, when, alongside your own CloudTrail, Azure Activity Log or Cloud Audit Logs.
AI agents: API tokens scoped by Open Policy Agent policies, so an agent gets its own identity and limited rights instead of an engineer's credentials (OPA policies in Qovery).
ID.AM-02: one view of clusters, environments and services per account, plus a Terraform provider if you want the inventory to come from code.
PR.PS-02: managed cluster upgrades on Qovery-managed clusters, which keeps Kubernetes versions inside support.
Preview environments per pull request, cloned from a template, so you can keep production data out of them by design.
Qovery publishes its own security posture on qovery.com/security, which is what you will need for your GV.SC-07 sheet on us. It does not make you NIS2-compliant, does not hold ACN cloud qualification, and does not replace your board approval, your incident plan or your legal review.
How can a coding agent help me prepare the inventory before an audit?
An agent connected read-only to your deployment platform can draft the ID.AM-02 inventory and spot environments without owners, which saves a day of spreadsheet work. Keep it read-only for this job: an inventory does not need write access, and PR.AA-05 is about least privilege.
BASH|Install the Qovery Agent Skills for Claude Code, Cursor, Codex, OpenCode or Gemini CLI
curl -fsSL https://skill.qovery.com/install.sh | bash
# Or install them for this repository only
curl -fsSL https://skill.qovery.com/install.sh | bash -s -- --project
# Add the Qovery MCP Server to Claude Code over OAuth, read-only by default
claude mcp add --transport http qovery https://mcp.qovery.com/mcp --callback-port 4242
The install script asks you to approve it, and the MCP Server opens an OAuth login in your browser, acting with your own user permissions. Then ask in plain language, for example "list every Qovery environment in my organization with its cloud account, region and cluster version". Write mode exists (claude mcp add --transport http qovery "https://mcp.qovery.com/mcp?read_write=true" --callback-port 4242, plus write access enabled in the Qovery Console), but you do not need it for audit preparation. Start with the AI agent quickstart or the Claude Code integration.
Frequently asked questions
Italy's NIS2 baseline deadline passed on October 31 and ACN has moved into its inspection phase. What should software vendors hosting on a hyperscaler cloud check before an audit?
Check the evidence for the ACN baseline measures that sit on your side of the shared responsibility line: supplier and cloud service inventories (GV.SC-04, ID.AM-04), supplier risk assessments (GV.SC-07), individual accounts with MFA and least privilege (PR.AA-01, -03, -05), centralised logs with documented retention (PR.PS-04), tested backups (PR.DS-11), supported software versions (PR.PS-02), written service levels (DE.CM-01) and an incident plan covering the 24h and 72h notifications (RS.MA-01). Also confirm that your board approved the measures under Article 23 of Legislative Decree 138/2024. ACN dates the deadline to October 2026, 18 months after your inclusion notice.
Is 31 October 2026 the official Italian NIS2 deadline?
ACN's Determinazione 379907/2025 sets the deadline at 18 months from the date each entity received its inclusion notice, and ACN's website dates that to October 2026 for entities listed in 2025. The 31 October date is the one Italian practitioners use as a reference. Your exact date is on the PEC you received from ACN, and entities first listed in 2026 have until 31 July 2027.
Can ACN inspect an important entity without a reason?
No. Article 36(2) of Legislative Decree 138/2024 limits verification and inspection powers over important entities to cases where ACN acquires or receives evidence, indications or information suggesting a possible violation. Essential entities can be inspected ex ante, including random checks. A significant incident notification can itself be the trigger for an important entity.
Does using AWS, Azure or Google Cloud make my company NIS2-compliant in Italy?
No. The hyperscaler secures its own infrastructure, while identities, MFA, logs, backups, patching, supplier management and incident plans stay with you under the shared responsibility model. ACN also expects you to list the cloud provider in your supplier inventory and assess it. If you run on IaaS or managed Kubernetes, you are the one who notifies CSIRT Italia when a provider incident hits your service.
What is the difference between ACN cloud qualification and NIS2 compliance?
ACN cloud qualification decides which cloud infrastructure and SaaS services an Italian public administration may buy, through the ACN catalogue and its ordinary, critical and strategic levels. NIS2 compliance under Legislative Decree 138/2024 applies to essential and important entities in many sectors and covers security measures, incident notification and board accountability. A vendor selling to Italian public bodies may need both, and holding one does not give you the other.
What are the fines for NIS2 non-compliance in Italy?
Article 38 of Legislative Decree 138/2024 sets fines of up to €10 million or 2% of worldwide annual turnover for private essential entities, and up to €7 million or 1.4% for private important entities, whichever is higher. Minimums are one twentieth and one thirtieth of the maximum. Essential public administrations face fines between €25,000 and €125,000, and Article 23 makes management bodies responsible for violations.
Morgan co-founded Qovery and leads engineering. He writes about Kubernetes architecture, DevOps best practices, and building resilient infrastructure at scale.
Next step
Get your deployment evidence ready before ACN asks for it.
Qovery deploys your applications into your own AWS, GCP, Azure or Scaleway account, or your existing Kubernetes cluster, with per-environment RBAC and an audit log of every change. Talk to us about mapping it to your NIS2 evidence pack.