AIエージェントが機器を安全に操作するには？ Anthropicが示した枠組み
2026-09-04
WIRED.jp
The information displayed in the AIM should not be reported as representing the official views of the OECD or of its member countries.
A malicious campaign named 's1ngularity' exploited compromised AI agents (e.g., Claude Code, Gemini CLI, Amazon Q) on developers' machines in Japan. Attackers remotely controlled these AI agents to enumerate and exfiltrate sensitive files, violating intellectual property and privacy by uploading data to public GitHub repositories without authorization.[AI generated]
Why's our monitor labelling this an incident or hazard?
The event involves an AI system explicitly mentioned as AI agents (e.g., Claude Code, Gemini CLI, Amazon Q) that are remotely controlled by malicious scripts to perform unauthorized data enumeration and exfiltration. The harm includes violation of intellectual property rights and privacy through unauthorized data exposure and breach of confidentiality, fulfilling the criteria for harm to property, communities, or violation of rights. The AI system's misuse directly leads to these harms. The article also discusses mitigation measures, but the primary focus is on the realized harm from the attack, not just responses or potential risks. Hence, the classification is AI Incident.[AI generated]