Hackers are embedding AI-generated malware hidden inside seemingly benign panda photos to covertly hijack Linux machines for cryptomining, based on Aqua Safety. The stealthy code evades antivirus software program and leaves nearly no hint.
The marketing campaign makes use of a “new breed of persistent malware,” combining image-based payload supply, AI-assisted scripting, and stealth methods equivalent to rootkit modules to keep up long-term management over contaminated methods.
Malware inside innocent-looking panda photos
Assaf Morag, Director of Risk Intelligence at Aqua Nautilus, recognized the malware as Koske — a modular menace designed to run quietly within the background. It’s able to mining a number of cryptocurrencies, adapting its conduct based mostly on the contaminated system’s configuration.
The malware is embedded in panda photos hosted on public image-sharing platforms. These photos, although seemingly innocent, comprise embedded code that infects Linux methods with out triggering conventional antivirus defences.
Aqua researchers traced the command-and-control infrastructure to a Serbian IP handle. The attackers initially gained entry by an uncovered JupyterLab occasion — a standard web-based interface utilized in knowledge science and growth workflows.
A warning of what’s to return
As soon as inside a system, Koske executes fully in reminiscence, compiling code on the fly reasonably than writing to disk. This in-memory execution helps it evade many widespread detection mechanisms.
A second payload contains software program that hides processes and recordsdata from view, or a rootkit. On this case, it makes use of LD_PRELOAD to hijack system features and make the malware invisible to primary monitoring instruments.
Components of the code present patterns typical of AI-generated scripting. Aqua researchers famous clear construction, modular logic, and neutralized syntax — that are hallmarks of huge language mannequin (LLM) involvement.
The malware is constructed to adapt. If one connection fails or a mining pool goes down, it switches to a different, utilizing public proxy lists and diagnostic instruments to maintain itself operating with out interruption. In keeping with Morag, “It’s a warning of what’s to return.”
AI’s rising position in cyber assaults
Koske is a part of a broader pattern involving AI-assisted cyber threats. Current instances embody deepfake scams focusing on firm executives and chatbots getting used to generate malicious code. In keeping with Verify Level, cyber assaults surged by 47% within the first quarter of 2025, pushed partly by automated toolkits and AI-generated malware that decrease the barrier of entry for less-skilled attackers.
Cryptocurrency stays a high goal. Chainalysis’ mid-year replace revealed that greater than $2.17 billion in crypto has been stolen to this point in 2025, with almost 1 / 4 of that tied to private pockets compromises. Specialists level to the accessibility of AI instruments as a key enabler of extra focused assaults on particular person customers.
Detecting malware that hides in plain sight
Aqua Nautilus urges customers to stay vigilant for refined system modifications that will point out hidden threats. These embody unauthorized modifications to .bashrc recordsdata and surprising background duties added by cron or systemd.
Locked modifications to DNS settings, equivalent to a modified /and many others/resolv.conf, can sign makes an attempt to regulate outbound site visitors. Sudden spikes in CPU or GPU use can also counsel cryptomining in progress.
Picture recordsdata or binaries compiled throughout runtime must be handled with warning as these might carry hidden payloads disguised as legit recordsdata, per the Aqua staff.
The researchers emphasised that scripting patterns with clear construction, modular logic, and generic feedback might point out AI-assisted malware. Moreover, community exercise involving instruments like curl or wget might reveal communication with distant attacker infrastructure.
These seemingly minor indicators, when seen collectively, level to a menace designed to remain hidden in plain sight.
Attackers are more and more utilizing AI to impersonate trusted platforms and bypass human defenses. Learn our protection on AI-generated phishing websites mimicking Okta and Microsoft 365 to be taught extra.
