Infrastructure Security for Government Cloud and AI

Cloud and AI move infrastructure security risk rather than remove it. Zero Trust, access control and monitoring that hold up, and the SolarWinds lesson.

What you need to know in 60 seconds

Government transformation is moving services to the cloud and adding AI, and none of it makes infrastructure security any less important. It moves the risk: away from the server room and towards identities, configuration, management tools and the suppliers who run them.

The threat is not easing. In October 2025 the NCSC said it was dealing, on average, with four nationally significant incidents a week[1]. Inside government, the National Audit Office found significant gaps in cyber resilience across 58 critical IT systems assessed in 2024, with at least 228 legacy systems still in use[2].

This article covers what infrastructure security means in a cloud and AI estate, what SolarWinds taught us about the tools that watch everything, how Zero Trust and network access control fit together, and where a security team should start.

What does infrastructure security mean once services move to cloud and AI?

Infrastructure security is protecting the layer everything else runs on: networks, identities, devices, platforms and the tools used to run them. In a data centre most of that was yours to see and touch. In the cloud, the provider secures its part, and the NCSC's cloud security principles exist to help you judge whether a provider meets your security needs[3].

Your share of cloud security does not shrink. Identity and access, configuration, secure administration and how your teams use the service stay with you, and no provider's certificate covers them on your behalf.

AI adds another layer. A model and the data behind it are assets with threats of their own. The government's Code of Practice for the Cyber Security of AI names data poisoning and indirect prompt injection among the risks, and asks organisations to secure the APIs, models and data their AI systems depend on[4].

Why is the risk rising for government now?

Three pressures are arriving together.

Volume. The NCSC supported 429 cyber incidents in its 2025 review year, and nearly half were of national significance[1].

The estate. One in three cyber security roles in government was vacant or filled by temporary staff in 2023-24, and departments had no fully funded plans to fix around half of their legacy IT assets, 120 of 228[2].

Regulation. The Cyber Security and Resilience Bill brings medium and large managed service providers into scope, classes data centres as essential services, and expects an initial incident notification within 24 hours and a fuller report within 72[5]. Suppliers that run public sector infrastructure will be held to the same bar as the services they support.

What did SolarWinds teach about the tools that watch your network?

Monitoring and management platforms see everything and can reach everything. That makes them the most valuable foothold an attacker can get. The 2020 compromise of SolarWinds Orion showed it at scale, and the NCSC's advice at the time was direct: keep affected Orion instances behind firewalls that block internet access, both inbound and outbound[6].

The lasting lesson is not about one product. The tools you use to run infrastructure need your strongest controls:

Monitoring is a security control only if the monitoring platform is itself secure.

How do Zero Trust and network access control fit together?

The NCSC describes Zero Trust as an architectural approach where inherent trust in the network is removed[7]. Every request is checked against policy, using what you know about the user, the device and the service. One of its principles puts it bluntly: don't trust any network, including your own[7].

Zero Trust is a direction of travel, not a product, and most government estates will run a mix for years. Cloud services can check identity on every request. Offices, prisons, courts and data centres are still full of devices that plug into a network. That is where network access control earns its place. A platform such as Cisco Identity Services Engine decides what each user and device may reach when it connects, using role-based policies, and keeps an unknown or unhealthy device away from anything sensitive. It turns the principle of knowing your user, service and device identities into something enforced at the point of connection.

What does secure by design look like for AI workloads?

Treat an AI service as infrastructure from the first day. The AI cyber security code of practice gives a practical starting list[4]:

None of this needs a separate security programme. It needs AI workloads to go through the same architecture review, access control and monitoring as everything else, with a few AI-specific additions.

What did this look like at the Ministry of Justice?

Justice Digital runs a high-security network estate across prisons, courts and headquarters, on PSN and CJSM connections. Its SolarWinds monitoring estate had capacity limits, duplicated nodes and no modern hardening, and its Cisco ISE policies needed to be live before the service could be brought fully in-house.

A Mayfair squad of SC cleared engineers audited and rebuilt the SolarWinds estate on a hardened, high-availability architecture covering 12,000 network and server devices. Automated discovery scripts and Ansible and Terraform pipelines brought on around 400 devices a day, with runbooks and a phased cutover synchronised with the outgoing managed service provider to keep monitoring running. In parallel, the team designed the Cisco ISE architecture and wrote role-based policy sets for network access control. Read the Ministry of Justice case study.

Common questions about government infrastructure security

What is Zero Trust, in plain terms?

Never assume a request is safe because of where it comes from. Each time a user or service asks for access, check who they are, what device they are using and whether policy allows it, whether the request comes from the office network or the internet.

Does moving to the cloud make government services more secure?

It can, because large providers invest heavily in the parts they run. But identity, configuration, administration and data access stay with the department, and a badly configured cloud service can expose more than an old on-premises system ever did. Use the NCSC cloud security principles to choose providers, and keep your own controls just as strong.

Do we still need network access control if we are adopting Zero Trust?

In most government estates, yes. Offices, prisons, courts and data centres are full of devices that connect to a network rather than straight to a cloud service. Network access control is how Zero Trust thinking reaches those devices at the moment they connect.

What will the Cyber Security and Resilience Bill mean for suppliers?

As drafted, medium and large managed service providers come into scope and data centres become essential services. Incidents will need an initial notification within 24 hours and a fuller report within 72, so suppliers running public sector infrastructure should check that their detection and reporting can meet those times.

Where should a government security team start?

Start with the tools that hold the most power over your network. List every monitoring, management, backup and remote access platform, who can administer it, and what it can reach. Then test each one three ways: is it cut off from the internet, is privileged access time-limited and logged, and would you notice if it started behaving oddly?

From there, map user and device access against the NCSC's Zero Trust principles and the Cyber Assessment Framework. Our guide to the NCSC CAF explains its objectives and principles on one page.

Mayfair IT Consultancy's network security team does this work across central government with SC and DV cleared engineers, from SolarWinds and Cisco ISE builds to architecture reviews and incident response planning.

Related reading