Observed incident

Aoki infostealer adds AI assistant credentials to its targets

Reported

What happened

Researchers found recent Aoki malware samples collecting AI assistant credential and session files, creating a route to stolen developer identities, paid API usage and connected services.

Parties and reporting

AI tools and systems: Claude, OpenAI Codex, ChatGPT, Cursor, Codeium, Gemini, GitHub Copilot, JetBrains AI Assistant, Kiro

Outcome

Symantec published sample indicators and detection coverage. The report does not identify victims, the delivery route or subsequent misuse of stolen AI credentials.

Operating lesson

Protect local agent credentials like production secrets; revoke tokens and sessions after endpoint compromise and investigate linked services.

Evidence and limits

Observed malware capability, not a confirmed breach of every named product. Aoki existed earlier in 2026; October 15 dates the newly reported AI-targeting capability. Actor identity and victim counts are unknown.

Sources