CVE-2026-15978 sgl-project CVE debrief

Who should care

Organizations using SGLang, security teams, and administrators responsible for model deployments should be aware of this vulnerability and take necessary actions to mitigate it. They should verify SGLang configurations and consider implementing compensating controls to mitigate potential model weight exfiltration. Additionally, they should monitor SGLang systems for suspicious activity and review SGLang configurations to ensure proper setup and security.

Technical summary

SGLang contains a model weight exfiltration vulnerability when no API keys are configured, allowing remote attackers to trigger distributed weight broadcasting and data transfer to exfiltrate model weights. The vulnerability is caused by SGLang exposing two endpoints that allow a remote attacker to trigger distributed weight broadcasting using NCCL and then triggering data transfer. Attackers can exfiltrate all model weights by exploiting this vulnerability.

Defensive priority

Organizations using SGLang should verify their configurations and consider implementing compensating controls to mitigate potential model weight exfiltration.

Recommended defensive actions

• Verify SGLang configurations to ensure API keys are properly set up.
• Implement compensating controls to mitigate potential model weight exfiltration.
• Monitor SGLang systems for suspicious activity.
• Review SGLang configurations to ensure proper setup and security.
• Verify affected SGLang deployments exist in managed environments and assign an owner for follow-up.
• Plan vendor-supported updates or mitigations through normal change control where exposure is confirmed.

Evidence notes

The CVE record indicates a model weight exfiltration vulnerability in SGLang when no API keys are configured. The NVD entry is currently Awaiting Analysis. Organizations should verify SGLang configurations and consider implementing compensating controls to mitigate potential model weight exfiltration. The vulnerability allows remote attackers to trigger distributed weight broadcasting and data transfer to exfiltrate model weights. SGLang contains a model weight exfiltration vulnerability when no API keys are configured, as SGLang will expose two endpoints that allow a remote attacker to trigger distributed weight broadcasting using NCCL and then triggering data transfer, attackers can exfiltrate all model weights.

Official resources