Balancing AI innovation with security

Business Post

9 hours ago

In the rapidly evolving landscape of cloud and cybersecurity, artificial intelligence (AI) is both a powerful enabler and a formidable challenge. While AI has long been a quiet force in the backend of security operations, its increasingly agentic and public-facing applications are poised to reshape how organisations defend their digital assets.
Günter Bayer, chief information officer, at Stryve, said it was important to understand the dual nature of AI in security: its transformative potential in bolstering defences and the critical imperative of securely integrating AI into business operations.
The key insight: while AI revolutionises cyber defences, businesses must abandon no-cost tools that compromise data security and invest in enterprise-grade solutions to maintain control over their digital assets.
For years, AI has been an unsung hero in the security world, operating behind the scenes to enhance existing tools, Bayer said.
'AI has been around in the security space in the back end for a while, but it is still human-assisted. For example, we have used a filtering service for over ten years, and it has AI built into the backend,' he said.
This human-assisted AI has underpinned foundational security services like advanced filtering, where intelligent algorithms analyse vast amounts of data to identify and block threats. Such AI-powered services have become instrumental in protecting organisations from malicious content and unauthorised access for over a decade.
The quiet efficiency of these backend AI systems has allowed security professionals to operate more effectively, sifting through noise to pinpoint genuine threats.
This evolution signifies a shift from reactive, signature-based detection to proactive, predictive threat intelligence, where AI's ability to discern patterns and anomalies is paramount.
Agentic AI revolution
The current wave of AI integration is markedly different. We are now witnessing an explosion of so-called agentic AI applications, where AI models are increasingly interactive and directly accessible to users. This shift, while promising, introduces new complexities.
As AI models become more refined and adaptable, their potential to revolutionise security operations grows exponentially. AI-driven systems can analyse colossal datasets, identify emerging attack vectors, automate threat responses, and even predict potential vulnerabilities before they are exploited.
Imagine AI autonomously patching known exploits, detecting sophisticated phishing attempts by analysing behavioural anomalies, or orchestrating a comprehensive response to a cyberattack in real-time. This future, where AI functions as a hyper-efficient digital guardian, is rapidly approaching.
This transformative power comes with a significant caveat: the imperative to securely leverage AI. As Bayer states, 'in the cloud world, in relation to the people's data, it's always your responsibility; it doesn't matter who you are hosting with, AWS or Azure or whomever.'
This fundamental principle of data sovereignty and responsibility remains unchanged, even with the advent of advanced AI.
Hidden cost of free AI
The burgeoning trend of individuals and businesses leveraging 'free' AI services presents a particularly thorny security issue.
'What does free mean,' Bayer said. 'All it means is you don't pay with money. You do pay some other way, though, and that's not great. If you are in business, you should just pay.'
Recent incidents underscore this danger. One noted AI, despite robust security measures, recently faced a vulnerability where specific prompting could lead to data exfiltration. This highlights that even well-designed AI systems can be exploited if not used with extreme caution.
Programmers are one group susceptible to these risks, Bayer said. The temptation to input production code into free AI services for rapid debugging or code generation is high. However, this shortcut can inadvertently expose proprietary algorithms, trade secrets, or even critical vulnerabilities to third parties.
AI has been around in the security space in the back end for a while, but it is still human-assisted
'I know AI is helping programmers a lot but, many of them, they just want a quick fix, so they're putting production code into free services,' he said.
Furthermore, malicious actors are increasingly adept at 'bypassing the guardrails' of AI models, using clever prompts and techniques to extract sensitive information or generate harmful content. This 'prompt injection' and data mining by unauthorised parties represent a significant threat that organisations must actively mitigate.
Bayer's view on this is unequivocal: 'Don't use any free service. If you're a business, pay for it. If you want, say, Copilot, then pay for your own instance.'
This advice is not about stifling innovation but about exercising due diligence and prioritising robust security. For businesses, investing in enterprise-grade, paid AI solutions that offer dedicated instances, robust data privacy controls, and clear service level agreements is paramount.
This ensures that data remains within a controlled environment and that the responsibility for its security is clearly defined. While AI tools offer immense benefits in productivity and efficiency, their integration must be approached strategically: balancing innovation with unwavering security principles.
Today, AI's role in cybersecurity is at a crossroads. Its power to enhance defences, automate responses, and predict threats is undeniable. However, the secure adoption of AI across organisations is not merely a technical challenge but also a cultural and strategic one.
By embracing AI with a clear understanding of its risks and a commitment to secure implementation, businesses can harness its transformative power to build resilient and future-proof cyber defences.