Digital Trust & Assurance · Part 1

Defensive Driving: Why Zero Trust Needs Situational Awareness in the Age of AI

Zero Trust can change an access decision when the signals change. The harder question is whether your organisation can see what has changed in the first place. Why Zero Trust depends on situational awareness, what AI changes, and what boards should ask.

Chris Keay, CTO, CyberteqPublished Nov 2026Digital Trust & Assurance series, part 1

In brief

  • Zero Trust can grant, limit or revoke access as signals change. It cannot decide which signals matter, or notice changes it has not been set up to see. That remains the organisation's job.
  • Four kinds of change can weaken a security decision that was sound yesterday: supplier access, shared technology, AI agents and attacker behaviour.
  • AI changes the conditions on both sides. It can help attackers find and use weaknesses faster, and AI agents take on new roles inside business workflows.
  • The 2026 refrigeration incidents show a dependency beyond an organisation's direct control, and the value of monitoring the condition a service actually depends on.
  • Boards should ask: when a critical dependency or a credible threat changes, who notices, who can change the controls and how quickly can they act? A recent example is better evidence than a list of controls.

Why doesn't Zero Trust manage security by itself?

Zero Trust rejects implicit trust based on network location or asset ownership. In NIST's Zero Trust Architecture guidance, access to a resource is authorised for a session and limited to what is needed. Policy can use signals such as identity, device health, the resource requested and observed behaviour, and an enforcement point applies the decision. Monitoring can prompt reassessment or revocation as conditions change.

That makes Zero Trust adaptive, but only within the signals it receives and the policies it has been given. Someone has to decide which signals matter, where they come from and what should happen when they change. Business relationships, digital ecosystems and organisations change continuously, and so do the threats to them. Zero Trust is not something an organisation implements once and leaves to run.

Digital Trust & Assurance

Are we watching the road?

Zero Trust can change an access decision when the signals change. The organisation has to see what changed.

The vehicleZero Trust: verify the driver, check the vehicle, permit the route
DriverVehicleRoute
ChangesSupplier accessA provider takes over remote maintenance1
ChangesShared technologyA new way to exploit shared equipment2
ChangesAI agentsAn agent gains permission to act3
ChangesAttacker behaviourWeaknesses found and used faster4
The journeyThe business outcome: a payment, payroll, blood kept at the right temperature

See what changed. Reassess access and response.Who notices, who can change the controls, and how quickly?

Think of a business activity as a journey. Zero Trust verifies the driver, checks the vehicle and decides which route is permitted, and it can revise that decision as new information comes in. But someone still needs to know when a bridge closes, who maintains the diversion and whether it can carry the vehicle. That is situational awareness: knowing how the business operates today, what it depends on and how threats to it are changing, and then acting on what you learn.

What changes along the road?

Four kinds of change can make a security decision that was sound yesterday less effective today.

ChangeWhat it can look likeWhat to watch
Supplier accessA provider takes over remote maintenance or adds a new route for remote accessWho can connect, to what, and how that access is enforced, monitored and revoked
Shared technologyA weakness is found in software or equipment used across many services and organisationsWhere the component is deployed and which critical services rely on it
AI agentsAn agent gains permission to update records, call tools or take actionsWho owns the agent, what it can do and which actions need approval
Attacker behaviourA new technique or campaign targets your sector or a component you useThreat intelligence relevant to your critical services and their dependencies

Supplier change is common and often hard to see. In the UK Government's Cyber Security Breaches Survey 2025/26, 15% of businesses reported reviewing the cyber risks posed by immediate suppliers, and 6% reported reviewing their wider supply chain. Among large businesses, 48% reported reviewing immediate suppliers.

Shared technology can carry a single fault or compromise into many organisations at once. Microsoft estimated that a faulty CrowdStrike update affected 8.5 million Windows devices in 2024, with no attacker involved. Log4j and MOVEit showed how a shared component or service could expose organisations that had never thought of it as a dependency.

What did the 2026 refrigeration incidents show?

On 2 November 2026, temperature and defrost settings were altered at three French Blood Establishment (EFS) sites storing plasma and samples. The incident involved refrigeration equipment maintained by an external provider, which reportedly said the weakness related to a component from a third-party supplier. EFS monitoring and alarms detected the changes, and staff intervened before any blood products were lost. EFS reported no interruption to collection and distribution.

Weeks earlier, on the night of 11 to 12 October, a cyberattack disrupted refrigeration controls at around 700 French retail sites, causing food spoilage and some fires. France attributed that campaign to Shahid Kaveh, an offensive cyber unit it identifies as affiliated with Iran's Islamic Revolutionary Guard Corps. Iran denies involvement. The attribution covers the October retail campaign; the blood-service incident is a separate case. Further incidents were reported in Europe in early November. Refrigeration failures at US military commissaries in August were investigated as a possible cyberattack, and on 5 November US agencies warned of an ongoing threat to refrigeration systems.

CERT-FR reported compromises of internet-exposed refrigeration controllers and advised operators to remove direct internet exposure, restore remote access only through secure connections and change default passwords. Public reporting has not confirmed the exact entry route, or whether AI played any role.

Two lessons stand out. First, the blood service depended on a system whose security sat partly with organisations beyond its direct control. Second, the response worked because staff were monitoring the condition the service actually depended on: the temperature of the stored products. Situational awareness is not only about the IT estate. It is about the outcome the business needs.

How is AI changing the conditions?

AI changes the conditions on both sides: for attackers, and for the business being defended.

Faster discovery and exploitation

AI can help attackers find weaknesses and develop exploits. Two published measures show how short the window has become, although they measure different things.

SourceWhat it measuredResult
Mandiant's 2023 analysisMean time to exploitation relative to patch availability, among vulnerabilities it tracked5 days after excluding 15 outliers; 47 days with them included
Microsoft's 2026 Digital Defense ReportMedian time from vulnerability discovery in the wild to weaponisationWell under 24 hours

The pace of AI

Exploitation timelines have shortened sharply

Mean time to exploit (Mandiant)

Days, relative to patch availability

* 2023 excludes 15 outliers; with them included, the mean is 47 days.

< 24 hours

median time from vulnerability discovery in the wild to weaponisation

Microsoft Digital Defense Report 2026

The two measures start and end at different points, so they cannot show on their own how much AI has changed exploitation timelines.

Sources: Mandiant / Google Threat Intelligence, Time-to-Exploit Trends 2023 (October 2024); Microsoft Digital Defense Report 2026 (October 2026).

The measures use different starting points, endpoints and statistics, so they cannot quantify AI's contribution or set a response deadline for a particular organisation. Microsoft's report notes that AI “can make a weakness easier to discover, an attack easier to customize, and a campaign easier to scale.” The practical implication is to know where a vulnerable component is used, which services rely on it, and how you can contain the exposure while a fix is tested or deployed. Mandiant's methodology explains its measure and outliers.

Agents with new roles

AI also changes the business being defended. An agent that summarises invoices has one set of risks. Give it permission to update supplier records and it can now affect where money goes. That change may happen in a business workflow long after the original access policy was agreed.

Zero Trust can help govern the agent's identity, permissions and access to tools, in the same way CISA's Zero Trust Maturity Model treats other identities. Situational awareness means knowing that the agent has taken on a new role. Adapting means deciding whether changes need approval, what to monitor and how to suspend the agent's access if something goes wrong. The OWASP Top 10 for Agentic Applications catalogues related risks, including agents that treat instructions embedded in a document as commands and misuse a tool.

What should prompt a review?

Start with your own journeys: the critical services you provide and the services you depend on. Follow them far enough to see where responsibility passes to a supplier, a shared platform or an agent. Then agree which changes should trigger a review.

When this changesReview
A supplier introduces a new route for remote accessWho can connect, what they can reach and what you can see
An agent gains permission to take an actionHow the action is approved, monitored, stopped and reversed
A new way to exploit shared equipment emergesWhere it is deployed and how quickly the exposure can be contained
A credible threat targets your sector or a dependencyWhich safeguards would still work, and who would act

These triggers only work if someone owns them, and if the information reaches people with the authority to change the controls.

What should boards ask?

For boards, the question is practical: when a critical dependency or a credible threat changes, who notices, who can change the controls and how quickly can they act? Ask for a recent example. It will tell you more about how well the organisation adapts than a list of the controls it has installed.

  1. Which critical services do we provide and depend on, and where does responsibility pass to a supplier, a shared platform or an AI agent?
  2. How would we know if one of those arrangements changed?
  3. Who has the authority to change access, isolate a system or suspend an agent, and how quickly can they act?
  4. Which threats are most pertinent to those services, and how does that intelligence reach the people who act on it?
  5. What happened the last time a dependency or threat changed, and how long did we take to adapt?
  6. How long can each critical service tolerate disruption, and what evidence shows we can recover in time?

Use the answers to decide which risks to reduce, avoid, accept or transfer. The next article in this series, We Cannot Eliminate Every Digital Risk, looks at that decision.

Where does cyber insurance fit?

A deliberate decision for each significant risk

More controls are not the only answer

ReduceApply or strengthen controls to lower likelihood or impact
AcceptRetain the risk knowingly, within agreed appetite, and review it
AvoidStop or redesign the activity that creates the risk
TransferShare the financial consequence through contracts or insurance, on an evidenced, shared view of the risk

Insurance may cover specified financial losses, depending on the policy. It does not notice change or restore a service. For state-backed attacks, cover depends on the policy's definitions, exclusions and attribution provisions; Lloyd's has required state-backed cyber attack exclusions in standalone cyber policies written in its market since March 2023 (Y5381, Y5433). Insurance works best when the client, broker, security provider and insurer share an evidenced, current view of the risk, rather than relying on an annual questionnaire.

Frequently asked questions

Is Zero Trust something you implement once?

No. Zero Trust adjusts access decisions within the signals and policies it has. Organisations need to keep deciding which signals matter, review policies as the business and its threats change, and act on what they learn.

What is situational awareness in cyber security?

Knowing how the business operates today, what its critical services depend on and how threats to them are changing, together with the ownership and authority to act when something changes.

What is the difference between Zero Trust and ZTNA?

Zero Trust is a security model and set of principles. Zero Trust Network Access (ZTNA) is one technology used to apply them, replacing broad VPN access with per-application connections granted on the basis of identity, device and context.

Is AI making cyberattacks faster?

Exploitation timelines have shortened sharply, and AI can help attackers find weaknesses and develop exploits. Available measures do not show how much of the change AI accounts for. AI agents also add new identities and permissions inside business workflows that need to be governed.

How should AI agents be governed under Zero Trust?

Treat each agent as an identity with an accountable owner, narrowly scoped permissions, limits enforced by the tools themselves, logging, and approval for consequential actions. Review the controls whenever the agent's role changes.

What happened in the 2026 refrigeration cyberattacks?

Refrigeration controls were disrupted at around 700 French retail sites on 11 to 12 October, which France attributes to an IRGC-affiliated unit and Iran denies. Further incidents followed in Europe, and on 2 November settings were altered at three French Blood Establishment sites, where alarms allowed staff to intervene before any blood products were lost.

Sources and further reading