You are probably already familiar with the World Economic Forum's report on cyber risks and AI. One of the statistics in the report that caught my attention was the finding that, in 2026, 64% of organizations have already begun conducting security audits of their AI tools. The question may arise: why has this become necessary? Today I would like to share my thoughts on this newly emerged necessity.
First, let me note that the report emphasizes that AI-related vulnerabilities are the fastest-growing cyber threat compared with the threats and risks that have existed worldwide to date. This in itself goes some way towards explaining why security audits of AI tools have become necessary.
Before turning to the threats, let me briefly outline the regulatory framework in this area in our country. As you know, by Order No. 530 of 19 March 2025, the President of the Republic of Azerbaijan approved the “Artificial Intelligence Strategy of the Republic of Azerbaijan for 2025–2028”.
The conclusion that countries around the world reached in 2026 regarding audits of AI solutions was reached by our country's President a year earlier: in the strategy I have just mentioned, conducting such security audits was assigned as a new responsibility of the Special Communication and Information Security State Service, which was tasked with issuing security opinions on AI solutions.
For information, it is worth noting that, alongside ongoing work in this area, the Special Communication and Information Security State Service has already drawn up security requirements for artificial intelligence solutions. These requirements will be submitted to the government for approval in the near future. In our country, where AI solutions are already being deployed at high speed, having such security requirements is important — and it is equally vital that the institutions and companies building AI solutions treat these requirements seriously and comply with them.
While on the subject of AI and security, I would also like to address several aspects of the issue that have not been discussed until now.
If one of them, as I mentioned earlier, concerned the security of AI solutions themselves, the other concerns AI tools and AI solutions themselves becoming targets for hackers.
One example of such targeting is the spread of the Djinn Stealer malware. Whereas hackers previously used stealer malware mainly to infect users' computers and seize the user data, passwords and accounts stored in their browsers, the new stealers now target the configuration files of AI coding assistants. Such attacks give hackers new opportunities to infiltrate AI development environments.
Another example is “prompt injection” operations targeting AI agents. Let me tell you about the most interesting of these: a new and unique attack campaign discovered by ZSCALER.
In this campaign, autonomous AI agents that browse websites, carry out analysis and make investments are manipulated through “prompt injection” planted on websites by hackers. AI agents manipulated in this way come to trust the hackers' fake platforms and, acting on the investment recommendations those platforms provide, are induced to make payments into the hackers' cryptocurrency wallets.
Another interesting targeted attack by hackers against AI solutions is the AI “HalluSquatting” attack. As you know, AI models can hallucinate. Against the backdrop of AI assistants “recommending” non-existent libraries while hallucinating, hackers have begun creating fake libraries under those names and distributing malicious packages designed to exploit these hallucinations.
The fact that this type of attack is so new and unique compels us to think deeply — and with concern — about new security vectors in AI security that have not been discussed until now.
The threats I have described are not limited to being new AI cyber threats. We must also recognize the importance of building new security environments, and keeping new risk matrices in view, for the AI solutions we will use in automated business tools and decision-making processes.
The results of a number of international surveys conducted today confirm what I am saying. If we look at the results of the survey conducted as part of Infosecurity Europe 2026, we see that while 87% of participating professionals confirmed that AI increases productivity, a quarter of them consider AI agents the greatest threat to organizations — a fact that shows just how great the risks to be taken into account are.
Companies are now obliged to protect not only people and their cyber-hygiene behaviour, but also the “trusted AI identities” that make business decisions.
Finally, let me draw your attention to one more interesting report. A report published by Secureframe states that 85% of defence industry companies are concerned that they will face AI-driven cyberattacks and deepfake social engineering within the next two years. For me, the more worrying finding is that, according to the report, 72% of companies are not confident that they can detect threats of this national-level kind.
Today I have spoken about a number of new threat vectors related to AI solutions. I will discuss the threats arising from the use of AI tools, the challenges awaiting organizations and AI data exfiltration separately in one of my upcoming videos. Thank you for your attention. Stay safe and well.
Sources & links
