What CISOs should invest in for 2027, mapped to NIST CSF 2.0

For a decade the default answer to a new threat was a new tool. Many large firms now run dozens of overlapping products, with scattered telemetry and plenty of shelfware. The NCSC expects AI to make attacks more frequent and more intense through 2027, while firms adopt AI themselves at speed. The question for a CISO has moved from “how do we stop the next attack?” to “can we run safely, recover quickly and prove our biggest risks are under control?” That changes where the money should go.

1. Identity becomes the control plane

Identity is no longer just the perimeter. It decides whether every action in the firm is allowed. Non-human identities are multiplying: workloads, APIs, service accounts, containers and AI agents. An agent that can read the CRM, query a database and start a workflow is a privileged user in all but name.

The NCSC’s August 2026 guidance on agentic AI says “all agents should be assigned their own unique identity” and that, where possible, firms “should restrict the credentials available to the agent”. For people, it says that where services support passkeys, “they should generally be used”.

• Invest in: phishing-resistant sign-in (passkeys, FIDO2), privileged access with just-in-time rights, automated joiner-mover-leaver, workload and agent identity, removing hard-coded secrets, identity threat detection.
• NIST CSF 2.0: PR.AA, identity management, authentication and access control.

2. Fund recovery as seriously as prevention

Most programmes are still measured on prevention: patches applied, emails blocked, alerts raised. Those numbers do not answer the board’s real question: if an attack succeeds tomorrow, can we keep trading?

A disaster recovery test that restores a server is not the same as rebuilding a customer service after Active Directory, endpoints and admin tooling have all been compromised at once. The biggest cost of a major incident is usually the disruption, not the initial break-in.

• Invest in: immutable and isolated backups, clean-room recovery, automated rebuild of core infrastructure, critical service mapping, destructive-attack exercises with the business, crisis management.
• NIST CSF 2.0: RC.RP, incident recovery plan execution; RS.MA, incident management.

3. Make AI security a formal domain

The debate about whether staff may use generative AI is over. The work now is controlling copilots, in-house models, AI features inside SaaS, retrieval systems, agents and AI-written code. Software that can reason and call tools changes the attack surface, and traditional application security does not cover all of it.

A workable AI security architecture answers eight questions:

• Who can access or change models?
• What data can they ingest, keep or expose?
• What can agents do, and which APIs and credentials can they call?
• How is prompt injection handled?
• Which models, datasets and plugins are trusted?
• Is abnormal behaviour detected?
• Who owns each AI system?
• Which business decisions does it influence?

The DSIT AI Cyber Security Code of Practice sets a baseline, including that the permissions an AI system holds on other systems are “only provided as required for functionality and are risk assessed”. In our view, CISOs will be judged on how safely they let the firm adopt AI, not on how well they slow it down.

• NIST CSF 2.0: GV.OC and GV.RM for context and risk strategy; PR.AA and PR.DS for access and data.

4. Security engineering at machine speed

The NCSC expects the most significant AI development in cyber to be AI-assisted vulnerability research and exploit development, and says the time between disclosure and exploitation has already shrunk to days. A firm that takes two weeks to assess a vulnerability, a week to find the owner and two more to schedule a fix is losing that race.

The aim is to shrink the cycle of discover, understand, prioritise, fix and validate from weeks to hours wherever possible, without simply producing more alerts.

• Invest in: attack surface and exposure management, risk-based prioritisation, automated control validation, software composition and infrastructure-as-code scanning, automated remediation, AI-assisted SOC triage.
• NIST CSF 2.0: ID.RA, risk assessment; DE.CM, continuous monitoring; ID.IM, improvement.

5. Put data security back at the centre

AI systems run on data. Connect an assistant to SharePoint, email, CRM and source code, and files that were technically open but hard to find become one question away. Old permission problems turn into real exposure. The test: do we know where our sensitive data is, who can reach it, and which AI systems can read it?

• Invest in: discovery and classification, entitlement clean-up, data security posture management, data loss prevention, encryption and key management.
• NIST CSF 2.0: ID.AM, asset management; PR.DS, data security.

6. Rationalise before you add

More tools do not mean more security. Overlapping products bring duplicated telemetry, licence cost, integration work and controls that are bought but half deployed. Before any new purchase, ask whether you already own the capability, whether an existing platform could provide it, what it will replace and what measurable risk reduction it will deliver.

The aim is fewer strategic platforms, integrated more deeply and used more fully. That is not a case for a single vendor.

7. Supplier resilience beyond questionnaires

A large share of cyber risk sits with cloud providers, SaaS, managed service providers, software and payment suppliers and, increasingly, AI providers. Annual questionnaires give a snapshot. Only 15% of UK businesses review risk from immediate suppliers, and 6% from the wider supply chain.

Regulation is pushing the same way. The Cyber Security and Resilience Bill brings medium and large managed service providers into scope, lets regulators designate critical suppliers and introduces an initial incident notice within 24 hours and a fuller report at 72 hours. Its Lords Report stage is scheduled for 26 October 2026. For banks and building societies, PRA reporting under PS7/26 applies from 18 March 2027.

• Invest in: continuous supplier monitoring, fourth-party and concentration mapping, SBOMs, contract security clauses, joint incident exercises, exit plans.
• NIST CSF 2.0: GV.SC, cyber supply chain risk management.

8. Start post-quantum work now

The task is not urgent replacement. It is the discovery needed to replace safely. The NCSC asks firms to finish discovery and an initial plan by 2028, complete the highest-priority migration by 2031 and finish by 2035. NIST published its first three post-quantum standards, FIPS 203, 204 and 205, in August 2024.

Most firms cannot yet answer a basic question: where do we use cryptography? Start there.

• Invest in: cryptographic and certificate inventories, crypto-agility in new systems, supplier readiness checks, identifying long-lived sensitive data.
• NIST CSF 2.0: ID.AM and PR.DS.

People: change how the team spends its time

AI can already help with log analysis, detection writing, alert triage, code review and compliance evidence. The opportunity is to take low-value work off skilled people, not to replace them. An architect should not spend hours collecting evidence by hand. The most valuable people will be those who can work across security, automation, AI, cloud and business risk.

Measurement underpins everything

Boards get plenty of numbers: vulnerabilities, incidents, phishing test results, patch rates. Few of them say how much risk the firm carries. Compare “98.3% of endpoints have EDR” with “every system supporting our Tier 1 payment service has tested detection, containment and recovery”. The second ties spend to a business outcome.

The UK Cyber Governance Code of Practice expects boards to govern cyber risk like any other material risk. A NIST CSF 2.0 current and target profile is a practical way to show the board what each pound moves.

Where 2027 budgets go wrong

Four questions for your next board or ExCo

Where do you stand?

Phishing-resistant identity for privileged users and isolated, tested backups is a Level 3 (Managed) capability. Investment mapped to a CSF 2.0 target profile and measured against important services is Level 4 (Measured). Recovery, supplier and AI controls tested together and reported to the board in business terms is Level 5 (Optimised).

A TrustOps Framework Assessment scores you against NIST CSF 2.0 and gives your board a costed, sequenced plan. Or start with the free 15-minute assessment.

General information, current as of October 2026. Not legal or regulated compliance advice. Verify against current primary sources.