Securing AI is not enough: why resilience must be part of the conversation

The conversation around AI security has moved quickly. As organisations adopt frontier and agentic AI, attention has understandably focused on securing models, preventing misuse and managing new forms of cyber risk. But security alone is no longer enough.

Much of the debate has centred on securing AI itself. How do we prevent prompt injection or model manipulation? How do we retain oversight when agents begin making decisions autonomously? And how do we stop attackers from using AI to gain an advantage over defenders?

As AI becomes embedded in business operations, customer service, software development, security and decision-making, organisations are becoming increasingly dependent on a relatively small number of providers, cloud platforms and technology ecosystems.

The question is therefore no longer simply whether an AI system is secure. It is whether the organisation can continue to operate if that system fails, becomes unavailable, is compromised or produces an unexpected result.

A new concentration risk

Cyber resilience programmes have long considered dependencies on critical suppliers, cloud services and digital supply chains. AI introduces another layer of reliance.

Many organisations now use the same handful of foundation models and hyperscale cloud platforms. But a concentration of capability also creates a concentration of risk.

An outage at a major AI provider could disrupt far more than one organisation. A vulnerability in a widely used model, or a problem affecting a major AI ecosystem, could spread across sectors and supply chains.

The risk is not limited to technical failure or cyberattack. Geopolitical tensions, regulatory intervention, commercial disputes and supply chain disruption could all affect access to technologies on which businesses increasingly rely.

This makes AI concentration risk a board-level concern for organisations. It belongs within enterprise risk management and business continuity planning, not solely within the security team.

Building resilience from the outset

However, resilience cannot be bolted on after those systems have become essential.

Businesses have already learnt difficult lessons from cloud adoption, software supply chains and third-party dependencies. They should apply those lessons to AI now, beginning with a clear view of where it is being used and which critical processes depend on it.

Leaders need to ask what would happen if a key AI service suddenly became unavailable. Which operations would be affected? Are there practical human fallback procedures? How quickly could services be restored? Could the organisation switch provider or model if necessary? And does it understand the dependencies within the wider AI supply chain?

These questions may attract less attention than model safety or prompt injection, but they could prove equally important. Organisations need contingency plans that work in practice, rather than assumptions that a service will always be available.

Securing the agents we deploy

Agentic AI adds a further challenge. Unlike conventional tools, agents can take actions, interact with systems and trigger workflows. Constant human supervision may be unrealistic.

Security teams are rightly concerned about prompt injection, manipulation, unauthorised activity and agent-to-agent exploitation. But resilience also depends on maintaining visibility and control when autonomous systems act on an organisation’s behalf.

That means building in clear accountability, auditability and appropriate human oversight from the beginning. Agents should operate within defined boundaries, with permissions proportionate to the task and the ability for people to intervene when something goes wrong.

This is particularly important in government, healthcare and critical national infrastructure, where an unreliable or unavailable system can have consequences far beyond lost productivity.

Managing vulnerabilities at AI speed

AI should not be viewed only as a threat. It offers one of the greatest opportunities to strengthen cyber defence in decades. Security teams are already using it to improve threat detection, identify risks faster and help overstretched analysts respond more effectively.

At the same time, AI is changing the economics of vulnerability discovery. As systems become better at finding weaknesses in software and infrastructure at machine speed, traditional patching processes may struggle to keep up.

The challenge will not simply be discovering vulnerabilities. It will be containing the risk at the speed at which AI can uncover them.

Reducing the window of exposure will therefore become critical. Virtual patching, attack-surface reduction, risk-based prioritisation and layered compensating controls can help protect organisations while permanent fixes are developed, tested and deployed.

The Government’s ambition for the UK to lead in AI-enabled cyber defence is both necessary and achievable. But success will require more than securing models and protecting agents. Organisations must be able to withstand disruption, manage concentration risk and respond quickly to vulnerabilities discovered at unprecedented speed.

Security and resilience are two sides of the same coin. As AI becomes woven into the fabric of modern organisations, the most important question may not be whether your AI is secure. It may be whether your organisation can continue to operate when it is not.

Cyber Resilience Programme activities

techUK brings together key players across the cyber security sector to promote leading-edge UK capabilities, build networks and grow the sector. techUK members have the opportunity to network, share ideas and collaborate, enabling the industry as a whole to address common challenges and opportunities together. Visit the programme page here.

Upcoming events

Latest news and insights

Learn more and get involved

Cyber Resilience updates

Sign-up to get the latest updates and opportunities from our Cyber Resilience programme.

Meet the team

Jill Broom

Head of Cyber Resilience, techUK

Jill leads the techUK Cyber Resilience programme, having originally joined techUK in October 2020 as a Programme Manager for the Cyber and Central Government programmes. She is responsible for managing techUK's work across the cyber security ecosystem, bringing industry together with key stakeholders across the public and private sectors. Jill also provides the industry secretariat for the Cyber Growth Partnership, the industry and government conduit for supporting the growth of the sector. A key focus of her work is to strengthen the public–private partnership across cyber to support further development of UK cyber security and resilience policy.

Before joining techUK, Jill worked as a Senior Caseworker for an MP, advocating for local communities, businesses and individuals, so she is particularly committed to techUK’s vision of harnessing the power of technology to improve people’s lives. Jill is also an experienced editorial professional and has delivered copyediting and writing services for public-body and SME clients as well as publishers.

Email:: [email protected]
Website:: www.techuk.org/
LinkedIn:: https://www.linkedin.com/in/jill-broom-19aa824

Read lessmore

Annie Collings

Senior Programme Manager, Cyber Resilience, techUK

Annie is the Programme Manager for Cyber Resilience at techUK. She first joined as the Programme Manager for Cyber Security and Central Government in September 2023.

In her role, Annie supports the Cyber Security SME Forum, engaging regularly with key government and industry stakeholders to advance the growth and development of SMEs in the cyber sector. Annie also coordinates events, engages with policy makers and represents techUK at a number of cyber security events.

Before joining techUK, Annie was an Account Manager at a specialist healthcare agency, where she provided public affairs support to a wide range of medical technology clients. She also gained experience as an intern in both an MP’s constituency office and with the Association of Independent Professionals and the Self-Employed. Annie holds a degree in International Relations from Nottingham Trent University.

Email:: [email protected]
Twitter:: anniecollings24
LinkedIn:: https://www.linkedin.com/in/annie-collings-270150158/

Read lessmore

Olivia Staples

Junior Programme Manager - Cyber Resilience, techUK

Olivia Staples joined techUK in May 2025 as a Junior Programme Manager in the Cyber Resilience team.

She supports the programs mission to promote cyber resilience by engaging key commercial and government stakeholders to shape the cyber resilience policy towards increased security and industry growth. Olivia assists in member engagement, event facilitation and communications support.

Before joining techUK, Olivia gained experience in research, advocacy, and strategic communications across several international organisations. At the Munich Security Conference, she supported stakeholder engagement and contributed to strategic communications. She also worked closely with local and national government stakeholders in Spain and Italy, where she was involved in policy monitoring and advocacy for both public and private sector clients.

Olivia holds an MSc in Political Science (Comparative Politics and Conflict Studies) from the London School of Economics (LSE) and a BA in Spanish and Latin American Studies from University College London (UCL).

Outside of tech, Olivia enjoys volunteering with local charities and learning Norwegian.

Email:: [email protected]

Read lessmore

Securing AI is not enough: why resilience must be part of the conversation

A new concentration risk

Building resilience from the outset

Securing the agents we deploy

Managing vulnerabilities at AI speed