Threadlinqs IntelligenceStart free

Threat video1:00

UNC6993 LLMjacking Cloud AI Access: Detection & IOCs

Stolen cloud tokens and LiteLLM flaws let UNC6993 resell AI model access. 9 detection rules, 16 IOCs and a free Splunk SPL rule cover it.

Read the full threat intel for UNC6993 LLMjacking Cloud AI Access: Detection & IOCs →

Chapters

  1. 0:00The hookAbout $1,000,000 stolen AI charges.
  2. 0:04The storyStolen keys gave cloud access.
  3. 0:10Attack pathFirst they harvested GitHub API keys.
  4. 0:25AttributionAttributed to UNC6993.
  5. 0:31Indicators of compromise7 indicators to hunt: 2 domains, 4 URLs and 1 package name.
  6. 0:38MITRE ATT&CK techniques17 ATT&CK techniques, from resource development to impact.
  7. 0:42Remediation4 steps now: Revoke and rotate exposed IAM credentials, Service Account keys, GitHub.
  8. 0:50Detection rules9 detection rules, in SPL, KQL and Sigma.
Transcript

0:00About $1,000,000 stolen AI charges.

0:05Stolen keys gave cloud access.

0:07They resold the access.

0:11First they harvested GitHub API keys.

0:14Validated access, escalated privileges.

0:17Enabled premium LLM inference at victim cost.

0:20Resold access through LLM Mixer.

0:26Attributed to UNC6993. Confidence: low.

0:327 indicators to hunt: 2 domains, 4 URLs and 1 package name.

0:3817 ATT&CK techniques, from resource development to impact.

0:434 steps now: Revoke and rotate exposed IAM credentials, Service Account keys, GitHub.

0:519 detection rules, in SPL, KQL and Sigma.

0:56Threadlinqs. The unified security engineering platform.