Threat & Exposure

AI in Cybersecurity: Benefits and Risks

The UK NCSC expects AI to make cyber intrusions faster and more frequent to 2027. What that means for defenders, and where AI itself adds attack surface.

Two AI pathways distinguish faster threats from defensive assistance, with human review governing decisions.
In this article 8 sections
ShareLinkedInWhatsAppEmail

Artificial intelligence is changing both sides of cybersecurity. The useful question for a leadership team is not whether AI matters, but which of its effects need a decision now.

What the NCSC expects to 2027

In an assessment published on 7 May 2025, the UK National Cyber Security Centre judged that “AI will almost certainly continue to make elements of cyber intrusion operations more effective and efficient”, increasing the frequency and intensity of threats. NCSC: Impact of AI on cyber threat from now to 2027

Two further judgements stand out. “The time between disclosure and exploitation has shrunk to days and AI will almost certainly reduce this further.” And the proliferation of AI-enabled tools “will highly likely expand access to AI-enabled intrusion capability to an expanded range of state and non-state actors.”

The worse side: faster and more convincing attacks

For most organisations this shows up in familiar places rather than exotic ones. Phishing and impersonation become more fluent and more tailored. Weaknesses in internet-facing systems are found and used sooner after they are published. The attacks are not new; the margin for slow patching and informal verification is shrinking.

A new attack surface: your own AI

The same assessment notes that the growing incorporation of AI models and systems across the UK’s technology base “almost certainly presents an increased attack surface for adversaries to exploit”. The point applies well beyond the UK. Every AI assistant connected to company data, and every model built into a workflow, is a system with access, inputs and outputs that someone can try to misuse.

Treat AI tools like any other system that handles sensitive data: know where they are used, what they can reach, who approved them and how their outputs are checked before anyone acts on them.

The better side: where AI can help defenders

AI can also shorten defensive work: summarising alerts, grouping related events, drafting first-pass analysis and helping analysts search large volumes of logs. In our view the gain is real when it frees skilled people for judgement, and risky when it replaces judgement. Keep a person accountable for any decision to contain, disclose or close an incident.

The divide that matters most

The NCSC warns that “there will almost certainly be a digital divide between systems keeping pace with AI-enabled threats and a large proportion that are more vulnerable”. Which side an organisation lands on depends less on buying AI tools than on doing the basics quickly and consistently.

What to decide now

  • Shorten the time from a published vulnerability to a patched internet-facing system, and measure it.
  • Require independent verification for payment, access and account-change requests, however convincing they sound.
  • Keep an inventory of AI tools in use, the data each can reach and who owns it.
  • Decide where AI may assist your security team and where a named person must make the call.

Questions for the next leadership meeting

AI risk is easier to govern when it is framed as a few concrete questions rather than a technology debate:

  • Which AI tools are staff using today, including free ones, and what company data has been put into them?
  • Who approves a new AI tool, and what does that approval check?
  • How long does it currently take us to patch a critical internet-facing system, and is that getting faster or slower?
  • If a convincing voice or video asked for an urgent payment tomorrow, which control would stop it?

Clear answers to these will do more for resilience than any single AI product. Where the answers are unclear, that is the work to prioritise.

Sources and further guidance

Regulatory applicability depends on the organisation and the provisions in force. Check the linked primary sources before acting.

Related reading and capabilities

Frequently asked questions

Does AI create entirely new kinds of attack?

For most organisations the near-term effect is that existing techniques, such as phishing, impersonation and exploiting known weaknesses, become faster and more convincing. The NCSC also highlights AI systems themselves as a growing attack surface.

Should we buy AI-based security tools?

Only where they free skilled people for judgement and the outputs can be checked. They do not replace fast patching, strong verification and clear ownership of decisions.

How do we govern staff use of AI tools?

Keep an inventory of the tools in use, the data each can reach and who approved it, and check outputs before acting on them, as you would for any system that handles sensitive data.