Page 169 - Cyber Defense eMagazine Annual RSA Edition for 2024

crucial question: how can this technology be effectively harnessed without being blinded by a false sense
of safety?

Challenges of AI-driven cybersecurity

AI-based cyber defense solutions can be exceptional tools, but only if they are applied correctly, and are
selected appropriately. It is not a set-and-forget type of solution — there are critical considerations for
choosing a suitable option. High among these, is the task of selecting a trustworthy cybersecurity SaaS
provider that has adequately prepared the software through specific training procedures. Additionally, it
must check the criteria of receiving ongoing maintenance procedures that ensure its effectiveness amidst
changing conditions.

During my decade-long experience in protecting core banking infrastructures, I’ve observed that the
effectiveness of an AI-based cybersecurity solution completely relies upon its training on a meticulously
curated dataset. This allows machine learning algorithms to distinguish their objectives. In particular,
feeding AI a variety of correct and incorrect use cases is vital to ensure it adheres to a highly nuanced
cyber defense protocol. This is a complex process that requires considerable expertise. Practically
speaking, checking whether all training has been adequately implemented and the solution can ensure
proper defense lines are not within the means of most businesses.

Herein lies the vital importance of due diligence. Being able to distinguish an AI’s practical cybersecurity
capabilities requires deep know-how of its functionality, otherwise the acquisition is purely based on the
promises of a provider. Merely relying on the reputation and marketing of a provider can eventually result
in a security breach — it is not enough in today’s challenging digital climate. Theoretically, even if a SaaS
solution provides 99% safety, the remaining 1% gap is enough to warrant an attack, at which point it is
too late for consideration.

This also entails the risk of overinvesting in a complete, stand-alone package. Lured by the promise of
AI, many companies could find themselves unable to fully utilize the solution — even in its best iterations,
it will require qualified steering by the likes of a CISO. Simply having the latest and greatest AI on the
market does not mean it will automatically be useful or effectively applied. That said, complete and stand-
alone packages are simply not possible to develop, considering the current shortcomings of AI. At best,
one can expect a great tool, which also acts as a platform that its developers constantly oversee, update,
and improve in response to the latest threats.

The belief that AI cyber defense can be a stand-alone model stems from a modern misconception.
Culturally, AI has gained an overestimated reputation — the hope of its potential often dazes people from
clearly evaluating its current limitations. It gets credit for undue accolades, especially in light of examples
like the famous AI robot Sophia being able to hold fluent conversations. By seeing such interactions, it is
easy to attribute general intelligence qualities where there factually are none. This is another point why
AI cyber defense solutions can not be trusted with an autonomous role. They need guidance, clear tasks,
and properly set objectives.

169