IncidentAI-enabled attacks
LameHug: Russia's APT28 ships malware that asks an LLM what to run
HighCERT-UA
Ukraine's CERT-UA reported LAMEHUG, malware sent by phishing to Ukrainian security and defence bodies that sends text descriptions to a coding model on the Hugging Face API and runs the commands it gets back. CERT-UA linked it with moderate confidence to APT28, which it describes as controlled by Russian special services. Google tracks the same malware as PROMPTSTEAL.
Replay the attack step by step in the Learning Center
How it happened
- Phishing emails carried a ZIP with a program disguised as a PDF.
- Once running, the malware asked the model to write commands to profile the system and find documents.
- It ran what the model returned and sent collected files to the attackers.
Impact
Government security and defence bodies in Ukraine were targeted with the first publicly reported malware that uses an LLM to generate its commands.
The fix
CERT-UA published indicators of compromise and detection guidance.
Lessons
- Malware can outsource its logic to a public AI API; outbound calls to AI services from odd processes are worth alerting on.
- Phishing is still the way in: phishing-resistant MFA and attachment controls still matter.