Aoki malware targets AI coding keys as OpenCode confirms leak

The two incidents landed in the same week but differ in mechanism. Aoki is an info-stealer that runs on an infected PC and collects stored credentials and session files; Symantec's Threat Hunter Team said the malware, first confirmed in 2026, added AI coding tool targets in recent samples. OpenCode's problem was on the cloud side: the operator said an old system being retired was accessed without authorization between the night of Oct 16 and Oct 17, and it notified affected users by email on the morning of Oct 17 (US time).

What ties them together is what a leaked key enables. An attacker can use the victim's account, run up usage the victim pays for, or resell the credentials. OpenCode described the breach as related to a "large-scale distillation attack" — using one model's outputs to train another — though it said the actual use of the leaked keys and whether individual billing reports connect to that attack need to be checked separately.

Symantec's report does not say how many victims there were or how stolen AI credentials were actually misused, and OpenCode's stated scope is distinct from users' reports of unexpected charges. The practical takeaway is to know where keys and tokens are stored, limit what each key can do, and be able to revoke one when something looks wrong.