Description

The Computer Emergency Response Team of Ukraine (CERT-UA) has identified a new phishing campaign linked to the Russian state-sponsored group APT28 (also known as Fancy Bear). The campaign aims to deliver a Python-based malware named LAMEHUG. Discovered on July 10, 2025, the malware was distributed via emails sent from compromised accounts impersonating Ukrainian ministry officials, targeting high-level government entities. The emails contained a ZIP archive with three malicious files: “???????.pif,” “AI_generator_uncensored_Canvas_PRO_v0.9.exe,” and “image.py.” LAMEHUG notably integrates a large language model, Qwen2.5-Coder-32B-Instruct, to dynamically generate executable system commands based on simple textual descriptions. The use of LLMs marks a sophisticated evolution in cyber threats, with the malware leveraging Alibaba Cloud’s Qwen2.5-Coder model via the Hugging Face API. This enables attackers to issue flexible and stealthy system commands, such as file harvesting from common user directories and system information collection. Extracted data is exfiltrated through SFTP or HTTP POST to attacker-controlled infrastructure. Furthermore, APT28 has been linked to another tool, Authentic Antics, which captures Microsoft credentials using stealthy login prompts and OAuth token theft, avoiding detection by manipulating Microsoft API parameters and registry paths. To mitigate such advanced threats, organizations should enhance email filtering and monitor unusual API traffic, especially toward services like Hugging Face. Blocking or closely inspecting outbound traffic to cloud-hosted APIs, applying network segmentation, and enforcing strict application allow-listing can reduce exposure. Regular audits and user training on phishing recognition are also essential for preventing compromise through social engineering.