Article by Gareth Cox
The pace of development in the AI space is showing no sign of slowing and the technology is beginning to reshape industries at a never-before-seen pace.
However, in the rush to adopt AI, there can be a recurring blind spot: IT security isn’t keeping up. In fact, according to IBM[1], innovation takes precedence over security in almost 70% of organisations.
Naturally every organisation wants to move faster and scale larger, however having speed without security is like building a rocket with no heat shield. It’s fine until it isn’t.
Security struggling to keep up
The problem isn’t just speed – it’s complexity. AI systems like large language models (LLMs) bring with them unique security challenges that most organisations are simply not equipped to handle.
The challenge lies in the very nature of AI itself. Unlike traditional software, AI systems learn, adapt and evolve. This makes them powerful, but it also makes them unpredictable.
A vulnerability in an AI system isn’t static. It can grow, shift and manifest in ways that are difficult to anticipate.
Many organisations pour millions of dollars into AI-driven defences, only to fall victim to a vulnerability they didn’t know existed. The truth is that no amount of AI and automation can entirely replace the human perspective.
AI is fast, but it lacks intuition. It doesn’t understand context in the way a seasoned security professional does.
The cost of not keeping pace
If an organisation’s IT security doesn’t keep pace with innovation, the consequences ripple far beyond the technical realm. These failures might stem from direct attacks or latent issues within the system, but the result is the same: Critical operations grind to a halt, exposing the organisation to both immediate and long-term disruptions.
Unfortunately, the risks don’t stop there. As AI becomes integral to customer-facing interactions, the stakes for failure rise exponentially. Missteps, whether from security breaches or flawed outputs, can erode trust and damage brand credibility.
Adding to the pressure is the growing attention on AI from governments and regulatory bodies, which are tightening expectations around privacy, safety and accountability at a pace that outstrips the readiness of many organisations. Failing to secure AI systems opens the door to not only breaches but also fines, sanctions and compliance challenges.
The absence of security can also stifle the very innovation AI aims to accelerate. When teams are consumed by crises or caught in a reactive cycle of patching vulnerabilities, they lose the bandwidth to focus on advancing their capabilities.
Over time, this reactive approach creates a compounding effect, where the gap between innovation and security grows wider, leaving organisations perpetually behind.
Adopting a security-first approach
To secure innovation at speed, organisations need to make security an integral part of innovation. This can be achieved by taking a range of steps including:
- Make security part of the design process:Security cannot be an afterthought; it must sit at the core of innovation. By embedding security protocols and considerations into the earliest stages of designing and deploying AI systems, organisations can address vulnerabilities proactively. This includes leveraging adversarial training techniques to stress-test AI applications and defining robust trust boundaries that segregate sensitive data from less critical components.
- Keep humans in the loop:While AI excels at processing and identifying patterns in vast data sets, it lacks the intuition and judgment that humans bring to ambiguous or novel situations. Employing a ‘human in the loop’ approach can help ensure that edge cases, ethical dilemmas and unpredictable anomalies are evaluated by experts. This hybrid model not only reduces risks but also fosters a deeper understanding of AI outputs.
- Implement AI-specific practices:Traditional cybersecurity measures fall short when addressing AI-specific risks like model inversion attacks or data poisoning. Organisations must adopt tools and methodologies designed for these challenges, such as implementing machine learning bills of materials (ML-BOMs) for transparency in AI supply chains or using advanced monitoring to detect irregularities in training and inference phases. These practices ensure that AI-driven innovation is both secure and scalable.
- Nurture a culture of adaptation:AI evolves, and so do its associated risks. Organisations need to commit to ongoing education, regular audits and the flexibility to adapt their security practices as new challenges emerge.
Innovating without compromise
As an organisation innovates, it must acknowledge the gaps that come with this speed and work diligently to close them. However, it’s not about slowing down but rather about keeping pace in a way that’s thoughtful, strategic and secure.
By embedding security at the core of their AI strategies, organisations don’t just protect themselves from evolving threats – they set the standard for responsible innovation and position themselves to lead with confidence in a future shaped by intelligent technologies.
[1] https://www.ibm.com/thought-leadership/institute-business-value/en-us/report/securing-generative-ai
