AI Malware: Lowering the Bar for Attackers

Your receptionist gets an email that looks like a shipping notification from UPS. The attachment is a ZIP file that, when opened, installs software that locks your files and demands a ransom. It is the same kind of phishing attack that has been hitting businesses for years — but the code inside was partly written by an AI.

That is no longer an edge case. Kaspersky’s Global Research and Analysis Team reports that large language models are now deeply embedded in attacker workflows, from generating malicious code to crafting phishing campaigns and enabling malware that adapts in real time.

AI does not invent new crimes — it scales the old ones

Let us be clear about what has changed. AI is not conjuring entirely novel forms of attack out of thin air. What it is doing is removing the barriers that used to keep most cybercrime in the realm of dedicated, skilled attackers.

For years, writing even basic malware required someone who understood how to code in at least one programming language, how to compile it, and how to avoid triggering antivirus scanners. That is changing. Groups like FunkSec have been observed using LLMs to generate portions of Rust-based malware — tools that steal data, encrypt files, and manipulate running processes. In the 2025 Revenge Hotels campaign, attackers used language models to produce infector and downloader code with little more than a text prompt.

The practical effect is straightforward: where one skilled developer used to be needed for a single attack, now a mid-tier criminal can produce and iterate on malware in hours. AI allows threat actors to move faster, adapt their tactics on the fly, and scale their efforts without scaling their staffing. Defenders have to be right every time; attackers with AI only need to be right once.

A new trick: malware that writes itself at runtime

Some of the most concerning developments are not about volume but about adaptability. Ukraine’s CERT recently reported a malware strain called LameHug, attributed to the state-sponsored group APT28, that contains no static instructions at all. Instead, when the payload runs, it sends base64-encoded prompts to an open-source LLM via the Hugging Face API. The LLM returns system-appropriate commands for reconnaissance or data collection — tailored to whatever machine it lands on, generated on the spot.

That means the malware is not just evading signature-based detection by being polymorphic; it is actively deciding what to do next based on what it finds inside your network. A single sample can behave completely differently on a point-of-sale terminal versus a back-office accounting workstation.

The AI-as-a-service disguise

There is another angle that hits smaller organizations especially hard. Kaspersky tracked more than 33,000 attacks on small and medium-sized businesses between January and April 2026 in which malware was disguised as popular AI services — tools like fake versions of AI writing assistants, image generators, and productivity bots. The number was five times higher than the same period in 2025.

The playbook is predictable: a business owner sees a link promising access to ChatGPT or a free AI image generator, downloads what they think is a useful tool, and gets a payload instead. The lure of AI is doing the social engineering work that a fake invoice or shipping notice used to do.

What you should actually do

The fundamentals of cybersecurity have not changed, but the window for getting them right has closed. Here is what matters now:

Keep your software updated, automatically. AI accelerates the discovery of vulnerabilities on both sides — defenders and attackers. The gap between finding a bug and patching it is the opening a real attacker uses. Enable automatic updates on everything you can, and treat manual update checking as a legacy habit.

Train people to be suspicious of “free AI” offers. An unexpected AI tool invitation arriving by email, Slack, or a text message is almost always hostile. If your team needs AI tools, provision them from known, trusted sources — never accept “free access” from a pop-up or unsolicited link.

Verify anything that asks for admin rights. Whether it is a software installer, a browser extension, or a remote-access tool, anything requesting elevated permissions should trigger a double-check with your IT provider first. Most modern malware still needs those rights to do real damage.

Make backups that are truly offline. Ransomware works the same way it always has — it only succeeds when there is one copy of your data and the attacker holds the key. Keep at least one backup that is disconnected from your network and your admin accounts.

Get professional eyes on your risk. AI has lowered the entry cost for attackers, but the defensive basics are the same. If you are not certain your updates, backups, and access controls are in order, book a free 15-minute consultation and we will make sure the things that matter are actually happening on your network.