Security For The Systems Your Enterprise Stack Cannot See

Industrial controllers, connected products and AI models now sit inside regulated business processes, and none of them accept an endpoint agent. IrisInfosec secures them on their own terms: IoT and OT security built to IEC 62443 and the EU Cyber Resilience Act, and AI security tested against the OWASP LLM Top 10 and MITRE ATLAS.

2

Disciplines under one accountable team: connected operations and AI systems.

62443

OT designed to IEC 62443 zones, conduits and security levels.

24h

The CRA early warning clock for an actively exploited vulnerability.

10

AI systems tested against the OWASP Top 10 for LLM Applications.

The IrisInfosec position

The Estate You Cannot Patch Is Still the Estate You Must Prove

Most security programmes assume a laptop: an agent installs, a patch lands in a maintenance window, a log reaches the SIEM. Industrial controllers, connected products and AI models break all three assumptions, and regulators have stopped treating that as an excuse. NIS2 asks operators to secure industrial networks. The Cyber Resilience Act asks manufacturers to secure what they ship and to report what gets exploited. The EU AI Act asks providers to make high-risk systems robust against manipulation. We design inside those constraints first, then produce the evidence in the shape the regulation asks for. Governance sits alongside in cyber risk and compliance, and proof in independent testing.

Two disciplines, one accountable team

What We Cover

Engage one discipline or run both as a single programme. Each links to a page with full scope, method and deliverables.

IoT & Security

Secure what runs the plant and what you ship. ICS and SCADA assessment to IEC 62443, industrial protocol and firmware analysis, and Cyber Resilience Act readiness for products with digital elements.

IEC 62443 EU CRA ICS / SCADA
Explore

AI Security & GenAI Protection

Test the model, not just the application around it. LLM and agent red teaming against the OWASP LLM Top 10, adversarial testing for poisoning and extraction, and security across the AI supply chain.

LLM red teaming MITRE ATLAS MLOps security
Explore

Not Sure Where To Start?

A short scoping call reads your industrial estate or your AI systems against the duties that apply, and tells you which work comes first.

Scoping call Posture review Priorities
Book a scoping call

How we work

From An Estate Nobody Fully Mapped To Controls
You Can Operate

A measured five-stage approach that respects what the environment can tolerate and ends in evidence a regulator will accept.

01

Discover The Estate

Inventory the controllers, devices and models actually in production, including what was deployed without a security review.

02

Model The Threat

Threat model to IEC 62443 zones and conduits, or to MITRE ATLAS and the OWASP LLM Top 10, depending on the system.

03

Test Inside The Constraints

Passive and out-of-band methods first, active testing only where the process or the model tolerates it.

04

Engineer The Controls

Segmentation, secure update paths, monitoring and guardrails designed for an estate you cannot rebuild.

05

Evidence And Sustain

Produce the technical file, the reporting path and the review cadence the regulation expects.

Regulations & frameworks

Regulator-Aware Engineering, Framework-Aligned

We design and test against the standards your auditors, your notified body and your regulator actually use, so one piece of work answers the operational question and the compliance question together.

One estate, several regimes. A connected product sold into the EU can carry Cyber Resilience Act, Radio Equipment Directive and sector duties at once. We map them once, not three times.

Primary sources, not hearsay. We work from the regulation and the standard as written, and cite the article or clause that applies to your system.

Safety sets the rules of engagement. In operational environments, process safety and availability constrain the test plan. That is a design input, not an obstacle.

Regulation or framework What we align to
EU Cyber Resilience Act Product security duties, SBOM in technical documentation, and the vulnerability reporting clock
NIS2 Risk management and incident reporting duties for essential and important entities
IEC 62443 Zones, conduits and security levels for industrial automation and control systems
EU AI Act Accuracy, robustness and cybersecurity duties for high-risk AI systems
OWASP LLM Top 10 &
MITRE ATLAS
Test coverage for LLM applications and adversarial AI techniques
ETSI EN 303 645 & EN
18031
Consumer IoT baseline and the RED cybersecurity harmonised standards

What you receive

Evidence for the regulator, drawings for the engineer

Practical artefacts, written so the people who answer to a regulator and the people who run the systems both find them useful.

Asset And Mode Inventory

Every controller, connected device and AI system in production, with owner, exposure and criticality.

Threat Model And Risk Register

Zones and conduits for industrial estates, MITRE ATLAS techniques for AI systems.

Reproducible Test Findings

Evidence a vendor cannot argue with and an engineer can act on, ranked by exploitability and consequence.

Secure Architecture Design

Segmentation, brokered remote access, update paths and guardrails scoped to the estate you actually have.

Product Security Technical File

Cyber Resilience Act documentation, SBOM and a vulnerability handling process that survives an audit.

Monitoring and Reporting Runbook

Detection tuned for industrial and AI telemetry, and the incident reporting path each regime expects.

Turn hard-to-secure systems into evidence you can present

Tell us what you run, and what you are
required to prove

Share your industrial estate or your AI systems and the regimes you report against. We will send a
prioritised plan with dates, usually within 3 business days.

Why teams choose us

Built for systems that do not fit the standard playbook

Most firms sell either industrial engineering or security consulting. We hold the regulation and
the engineering together, so the design and the evidence come from the same team.

01

Regulation And Engineering In One Team

The people who read the Cyber Resilience Act are the people who write the test plan, so nothing is lost in translation.

02

Safety-Aware Testing

We assess live industrial estates without becoming the incident, using passive capture, lab replicas and agreed abort criteria.

03

Operator And Manufacturer View

We work with the plant that runs the equipment and the firm that builds it, so both sides of a duty are understood.

04

Multi-Jurisdiction Range

EU, US and India understood as peer markets, so one design carries across the regimes you sell and operate in.

05

Specialist, Not Generalist

Security and regulation are the whole job, so you get depth a broad consultancy struggles to match.

06

Certified Partner Ecosystem

Delivery is backed by an ISO/IEC 27001 certified partner network, so capacity does not limit your timeline.

Questions, answered

Advanced security domains FAQ's

Clear answers to what security and product leaders ask us most. Need something
specific to your sector? Ask us directly.

Advanced security domains covers the systems that standard enterprise security tooling was never built for: industrial control systems, connected products and AI models. They share a constraint. You usually cannot install an agent on them, you cannot patch them on demand, and their failure modes are physical or statistical rather than purely digital. They are secured by design, testing and evidence rather than by deployment.

No. IT security optimises for confidentiality first and tolerates a reboot. Operational technology optimises for safety and availability, runs equipment with lifecycles measured in decades, and treats an unplanned stop as the incident. The controls look different as a result: segmentation and brokered access instead of agents, passive monitoring instead of active scanning, and IEC 62443 zones and conduits instead of a flat asset list.

If you place a product with digital elements on the EU market, most likely yes, whether you are the manufacturer, the importer or the distributor. From 11 September 2026 manufacturers must report actively exploited vulnerabilities and severe incidents, with an early warning within 24 hours and a full notification within 72 hours. The wider obligations, including essential requirements, technical documentation and an SBOM, apply from 11 December 2027.

The Digital Omnibus on AI, in force since July 2026, moved the application dates for high-risk obligations under the EU AI Act. Stand-alone high-risk systems listed in Annex III move to 2 December 2027, and high-risk AI embedded in regulated products moves to 2 August 2028. The obligations themselves, including the accuracy, robustness and cybersecurity duty, did not go away. The preparation window moved.

Both, and the distinction matters. An application penetration test finds the flaws in the interface, the API and the session handling. It does not find prompt injection through a retrieved document, a poisoned training set, an over-permissioned agent or a model that leaks its training data under extraction. We test the model, the pipeline and the agent as well as the application.

Governance sets the rules, this pillar secures the systems that ordinary controls cannot reach, and independent testing proves the result. AI governance under cyber risk and compliance answers whether you may run a system and who is accountable for it. This pillar answers whether that system can be manipulated. Enterprise cyber defense covers the corporate network the industrial estate connects to.

Secure the systems that do not fit
the standard playbook

Book a call with an advisor who reads the regulation at source and has
stood in a plant. We will send a staged plan with dates, no obligation.