Anthropic warned that cybercriminals are rapidly operationalizing advanced AI, releasing a threat-intelligence report detailing how “agentic” systems now execute end-to-end attacks rather than merely advise on them. The company said one actor used Claude Code to automate reconnaissance, credential theft, data exfiltration and tailored extortion campaigns across at least 17 organizations, while another scheme saw North Korean operatives leveraging AI to fabricate identities, pass technical interviews and perform on-the-job tasks at U.S. tech firms—sidestepping long-standing sanctions. In a third case, a low-skilled developer reportedly used AI to build and sell ransomware variants with anti-analysis and evasion features.
Anthropic said it banned implicated accounts, rolled out specialized classifiers and new detection methods, and shared technical indicators with authorities. The company cautioned that AI is lowering barriers to sophisticated cybercrime and embedding itself across victim profiling, data analysis and monetization. Additional cases in the full report include an attempted compromise of Vietnamese telecom infrastructure and multi-agent fraud operations. With AI-enabled attacks growing more adaptive, Anthropic called for tighter defenses across industry and government.
Related articles:
NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0)
Europol Internet Organised Crime Threat Assessment (IOCTA) 2024
MITRE ATLAS: Adversarial Threat Landscape for Artificial-Intelligence Systems
