New AI-Driven Malware & Security Threats Emerge, Including PoeLLM Cryptomining Attacks

Key Intelligence

• ■A new malware strain, PoeLLM (also known as Mythos 5), has infected over 3,000 servers, primarily for cryptomining, by exploiting exposed AI infrastructure.
• ■PoeLLM utilizes sophisticated techniques, including hiding command-and-control (C2) addresses in GitHub poems and successfully bypassing CAPTCHAs to mimic human behavior.
• ■Beyond cryptomining, other AI-related threats include fake AI SDK packages installing Remote Access Trojans (RATs) and AI-powered penetration testing tools being repurposed for data theft against financial institutions.
• ■Experts highlight a critical disparity: AI-driven attacks can succeed within minutes, while traditional security measures like biannual penetration tests are insufficient to keep pace with these rapidly evolving threats.