Executive Summary
A new variant of the Aoki infostealer, first identified in early 2026, has shifted its focus from traditional browser and cryptocurrency data to targeting the credentials of AI coding assistants. According to reports from Symantec, the malware is now capable of harvesting API keys and session tokens from tools including Anthropic's Claude, OpenAI's Codex, Cursor, and Google’s Gemini.
Technically, the malware targets specific configuration and cookie files, such as .claude[.]json and various .env or config[.]toml files, to exfiltrate sensitive access data. By hijacking these sessions, attackers can operate under a developer's identity, consume API quotas, and potentially bypass security safeguards.
This development is significant as it indicates that AI-related credentials have become high-value targets for monetization and lateral movement. Stolen tokens can grant attackers access to connected source code repositories, cloud environments, and CI/CD pipelines, posing a severe risk to intellectual property and software supply chain integrity.