Threadlinqs IntelligenceStart free

Threat video1:00

NEAR AI Breach Through SGLang Flaw: Detection & IOCs

Attackers used an SGLang flaw to reach a retired NEAR AI Private Chat backend. Spot it with 9 detection rules, 9 IOCs and a free Splunk SPL rule.

Read the full threat intel for NEAR AI Breach Through SGLang Flaw: Detection & IOCs →

Chapters

  1. 0:00The hookNEAR AI breached: about 100 user accounts.
  2. 0:04The storyAttacker got into one inference server.
  3. 0:10Attack pathAn SGLang flaw gave an attacker a foothold.
  4. 0:28AttributionNo threat actor is attributed.
  5. 0:32Indicators of compromise4 indicators to hunt: 3 URLs and 1 package name.
  6. 0:38MITRE ATT&CK techniques5 ATT&CK techniques, from initial access to collection.
  7. 0:43Remediation4 steps now: Revoke all active sessions and administrative credentials on AI inference.
  8. 0:50Detection rules9 detection rules, in SPL, KQL and Sigma.
Transcript

0:00NEAR AI breached: about 100 user accounts.

0:05Attacker got into one inference server.

0:07Then moved to a retired feature's system.

0:11An SGLang flaw gave an attacker a foothold.

0:14Commands ran, then reached Private Chat.

0:17About 100 users' details downloaded.

0:1918 legacy agents' data was accessed.

0:22Two customers' API keys were in logs.

0:28No threat actor is attributed. Confidence: low.

0:334 indicators to hunt: 3 URLs and 1 package name.

0:385 ATT&CK techniques, from initial access to collection.

0:434 steps now: Revoke all active sessions and administrative credentials on AI inference.

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

0:56Threadlinqs. The unified security engineering platform.