AI Cyber Threats: What The NCSC's 3 November Assessment Means For Financial Firms Using AI

2028 - Horizon of the NCSC's assessment of AI-enabled cyber threats (NCSC, 3 November 2026)  ·  3 - Scenarios the assessment uses, from a baseline rise in threat to a slower pace of AI-cyber development  ·  5 - Key drivers the NCSC identified as likely to shape the AI-cyber threat to the UK in 2028

AI cyber threats are attacks in which criminals or states use AI to find weaknesses, write malicious code or move faster than defenders can respond. The NCSC's assessment, published on 3 November 2026, is the clearest official view of where that is heading in the UK. It applies to every organisation, and financial firms adopting AI face it from two sides: attackers using AI against them, and their own AI tools becoming something to attack.

What did the NCSC assess on 3 November?

The NCSC Assessment examines current AI-enabled threat trends and projects the threat to 2028 through three scenarios: a baseline in which the global threat from AI-cyber capability increases, a pessimistic case in which the spread of AI-cyber capability far outstrips the UK's defensive ability, and a case in which AI-cyber development slows and defenders catch up. The NCSC says the scenarios are not predictions but are designed to show possible paths.

• **Nation states.** By 2028 it is highly likely that Russia, China and other nation states will routinely use AI-cyber to achieve their geopolitical goals, with the UK facing an increase in the scale and speed of sophisticated attacks.
• **Vulnerable systems.** By 2028 it is almost certain that many systems will remain vulnerable to AI-cyber tools that identify and exploit system weaknesses.
• **Legacy and AI attack surface.** By 2028 it is likely that attackers will apply AI-cyber capability to known weaknesses in legacy technology in critical national infrastructure, and to the new attack surface created when AI tools are deployed.
• **Ransomware.** It is highly likely that between now and 2028 AI-powered malware will have been used in a ransomware attack on UK critical national infrastructure.

The assessment uses the Professional Heads of Intelligence Assessment probability yardstick, so terms such as highly likely and almost certain carry defined meanings.

Why do AI cyber threats matter for financial firms?

Finance is one of the UK's critical national infrastructure sectors, and smaller firms sit inside the same supply chains as the largest ones. Two points in the assessment matter most for firms with 10 to 250 staff. First, the NCSC says defensive AI tools have shortened the window between a weakness being discovered and being widely exploited, and have made patching quicker. Firms that patch slowly will stand out. Second, every AI tool you deploy is new attack surface. An agent with broad access and no approval step is exactly the kind of opportunity the NCSC describes.

How should firms deploy AI tools with AI cyber threats in mind?

• **One identity per AI tool.** No shared accounts, so you can see and switch off each tool on its own.
• **Read-only by default.** An agent that only reads cannot change records if it is misused.
• **A person on every decision.** Anything that changes a record, sends a message or moves money waits for a named person.
• **A log you own.** Every action recorded where the tool cannot edit it.
• **Patch on a timetable.** Include AI tools and their libraries in your patching cycle and supplier checks.

These are the same controls behind our line, “The agent prepares. A person approves.” Our Desk Twin is built on them: read-only connections, approvals before anything leaves and a log of every step. Our article on AI agent security gives six questions to ask any vendor.

What should a board ask about AI cyber threats?

• Which AI tools do we use, who owns each one and what can each one reach?
• How quickly do we patch, including AI tools and the software they depend on?
• If an AI tool were misused tomorrow, could we switch it off and show what it did?
• Which of our suppliers use AI on our data, and how do they control it?

Our earlier pieces on AI regulation in the UK and the sector's cloud outage exercise cover the regulatory and resilience side.

Where should a firm start?

List every AI tool and agent in use, with its owner and access, and close anything broader than it needs. Our AI Agency London team builds every agent with these controls from day one, as our services page shows, and a 14-day Proof Run puts one workflow live inside them. This article summarises a published NCSC assessment and is not security or regulatory advice.