Expert explains how to protect data when using AI
In the age of artificial intelligence (AI), businesses of all kinds are trying to leverage this new technology. But the other side of the coin is growing cybersecurity risks, as AI agents can be hijacked.
According to Murat Kenebayev, IT security advisor at Freedom Cloud Holding, every user and business should follow time-tested rules: use a strong password, enable two-factor authentication, and never open suspicious emails to minimize cybersecurity risks. However, this is just the bare minimum because today, the key target isn’t just the user, but the trust between users, services, applications, and AI.
This is especially challenging for small and medium-sized enterprises that have no specialized teams to monitor cyber threats around the clock. That said, reliable cyber protection often isn’t that complex, but it does have to be systematic.
«Some people might think that multi-factor authentication (MFA) is the ultimate solution. But that is an illusion — and a dangerous one,» the expert emphasized, noting that it is crucial to know which apps you’ve granted access to your email, files or other information.
A simple rule for any action involving money, bank account details or the transfer of sensitive information is never to use the same channel through which you received the request to confirm a critical action.
«If a request arrives via a messaging app, confirm it by phone. If it comes via email, confirm it through a corporate communication channel or a known phone number. Do not confirm a critical action simply by replying to the same message. A second, independent channel does not make a company invulnerable, but it makes life much harder for an attacker,» Kenebayev said.
Companies can also monitor not only who logs into their systems but what they do after logging in. For example, downloading thousands of files for no apparent reason is solid grounds for suspicion and could warrant an investigation.
«For me, the key principle is this: a company must see not only that access occurred, but also what happened as a result,» the expert added.
He believes there is no need to fear AI, but also no reason to let it manage everything. A human must be the final decision-maker in every scenario. The Open Worldwide Application Security Project (OWASP), for example, recommends restricting an agent’s access to only those tools and resources strictly necessary for its task, while keeping critical operations outside the model’s direct scope of authority. In other words, if the AI makes a mistake, that error should occur within predefined, safe boundaries.
Ultimately, the goal is not to eliminate the «human factor,» which is simply impossible, the expert argues, but to ensure a single mistake doesn’t turn into a catastrophe for the entire business. In this scenario, even a successful attack on a business has a limited impact — and that is a result worth striving for.
