Talent LinkedIn
SDP Deep Tech
Let's talk
Case 01
IndustrialEntity not identifiedIEC 62443NIS2

Real cases

A single incident could bring the whole plant to a halt

Problem

Machinery, engineering, production and offices had gradually been connected to the same network. A compromised computer could reach essential industrial devices directly. Legacy protocols, uninventoried assets and machines with no operational alternative.

Framework applied

Asset inventory, traffic flow analysis and a zone-segmented architecture in line with IEC 62443, restricting communication between areas without stopping production. NIS2 risk management and continuity principles where the activity and size require them.

Measured outcome
1 → 7
Distinct security zones
186 → 23
Expressly authorised communications
100 % → 1 zone
Potential reach of an incident
Case 02
GovernmentEntity not identifiedRD 311/2022GDPR

It worked, but its security could not be demonstrated

Problem

Electronic services live without the evidence to withstand an audit: incomplete categorisation, an outdated risk analysis, measures with no documentary backing and a supplier whose responsibilities were never delimited.

Framework applied

Architecture review, system categorisation, updated risk analysis and organisation of evidence in line with the Spanish National Security Framework, coordinated with the GDPR where personal data is involved.

Measured outcome
42 % → 94 %
Measures with verified evidence
31 → 3
Findings, from eight material ones to minor observations
0 → 2
Documented continuity exercises per year
Case 03
Private sectorEntity not identifiedENSContinuity

The report had come back green every night for three years

Problem

Not one full restore, ever. Backup credentials depending on the very environment they protected, repositories reachable from the corporate network, critical systems only partly covered, and recovery times nobody had measured.

Framework applied

Classification by criticality, one isolated copy, repository integrity checks and real restores in a controlled environment, with recovery order and owners documented. The ENS covers continuity and the capacity to respond and preserve.

Measured outcome
0 → 12
Annual restore tests
>48 h → 6 h 20 min
Validated recovery, previously only estimated
67 % → 100 %
Critical systems recoverable
Case 04
Cross-sectorEntity not identifiedRules of engagementENS · NIS2

They wanted to see what an attacker would see, before it happened

Problem

Portals, remote services and suppliers added without any view of the public exposure: disused services still published, forgotten subdomains, reachable admin panels and technical information that made preparing an attack easier.

Framework applied

With prior authorisation and a defined scope, assessment from the perspective of an external attacker. Only the necessary vulnerabilities are verified, evidence is preserved and availability is not affected. Findings aligned with ENS or NIS2 depending on scope.

Measured outcome
37 → 21
Public assets, from identified to necessary
11 → 0
Exploitable critical or high vulnerabilities
9 → 0
Services with no assigned owner
Case 05
Cross-sectorEntity not identifiedISO/IEC 27037Chain of custody

The device had failed; the information was still essential

Problem

An unstable medium with damaged sectors, earlier recovery attempts that may have altered the information, no initial forensic image and no record of who had handled it, with sensitive data mixed in with material outside the case.

Framework applied

Identification and registration of the medium, preservation of the original, a working image where technically possible, and documentation of every step, with analysis limited to what was authorised. Integrity, traceability and minimisation, with ISO/IEC 27037 as the reference.

Measured outcome
72 % → 96 %
Information recovered from the medium
18,400 → 1,260
Files recovered and within the authorised scope
0 → 100 %
Actions with a documented chain of custody
Case 06
RoboticsClient not identifiedRegulation EU 2023/1230CRA

The robot was ready; the infrastructure would not leave the lab

Problem

A prototype processing vision, sensors and models ran well on the bench but could not be deployed repeatably at a client site: models on development machines with no reproducible configuration, no separation between training and production, updates without validation or rollback, and communications with no security model.

Framework applied

Turnkey infrastructure with compute, model versioning, observability, protected communications and an update procedure, documented so it can be replicated. Where AI takes part in movement with physical impact, the regulatory route runs through the Machinery Regulation, not the AI Act’s high-risk regime.

Measured outcome
3 days → 2 h
Deployment per robot
14 → 100 %
Components and models versioned
91.4 % → 99.2 %
Availability measured during validation
Case 07
International mobilityEntity not identifiedNIS2ISO/IEC 27001

The executives travelled; security stayed at the office

Problem

Laptops and phones with access to email and business applications from different time zones and jurisdictions with different access frameworks. Alerts were only reviewed during office hours: an anomaly at 3 a.m. went unexamined for hours. Five uncorrelated sources and no way to contain a device out of hours.

Framework applied

Integrated monitoring of identities, endpoints, mobile devices, email and cloud under a 24×7 model, with use cases for atypical access and credential compromise, plus a travel protocol with least privilege and remote isolation. The criterion is each profile's mobility, exposure and criticality, whatever the destination.

Measured outcome
8×5 → 24×7×365
Monitoring and response capability
>10 h → <15 min
Initial triage of an alert
5 → 1
Alert sources, from independent to correlated
0 → 100 %
High-mobility devices under a hardened protocol
Case 08
Cross-sectorEntity not identifiedRegulation EU 2024/1689GDPR

The board was still deciding how to adopt it; the data was already going in

Problem

Middle managers using personal generative AI accounts with contracts and internal documentation. In parallel, an industrial pilot contracted without assessment or an assigned owner. No inventory, no criteria for classifying information and no procedure to approve or withdraw a system. AI had arrived ahead of its governance.

Framework applied

A diagnosis of the AI in plain sight and the AI that was not, classification by purpose and impact, defined corporate tools, an owner per system, assessment before any procurement, human oversight and training. The Spanish Data Protection Agency warns that a prompt may contain personal data or trade secrets, and that entering them may constitute processing and a disclosure to a third party.

Measured outcome
0 → 14
AI systems inventoried and classified
9 → 0
Personal accounts used with corporate information
0 → 100 %
Systems with owner, purpose and approval status
18 → 100 %
Users trained and covered by policy

We never publish architectures, vendors, versions or console captures: showing a client’s blueprint would itself be a security incident.

Every case corresponds to real work. Figures and details have been minimised and anonymised in line with the confidentiality agreements signed, so that no organisation is identifiable.