VulnSocial

CVE-2026-15978Disclosure(lmsys / sglang)

LOWCVSS 7.5 · HIGH

Signal is active with 3 mentions in latest observed window

Immediate actions

  • Track advisory updates for patch or workaround availability

Recommended action window: Monitor and triage in normal cycle

NVD description

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.

lmsys CVEsBack to dashboard
0.0/ 10 priority

Sources & remediation

Vendor / third-party advisories
Other references
Weakness type (CWE)
CWE-306

Priority

LOW

Exploitation

NONE

PoC

NONE

Patch

NONE

Momentum

NONE

Are you affected?

If you run products in this scope, you should treat this CVE as relevant to your environment.

  • sglang

Threat summary

  • 3 mentions across 1 observed day

What's happening

  • Technical details provided in 3 signals
  • Disclosure: 3 classified signals
  • 3 total mentions across 1 day

Affected systems

Vendors
lmsys
Products
sglang

Deep dive

Activity timeline3 mentions / 1d
01223Mentions · 2026-07-30: 3Technical Details · 2026-07-30: 307-30
Signal classification1 categories
Disclosure
3100.0%
Referenced assets3 URLs
Full discourse3 posts
  • OJOBIT@0J0BIT
    Disclosure

    CERT/CC disclosed six vulnerabilities in SGLang, an open-source LLM serving framework > CERT/CC published VU#281278 detailing six distinct CVEs: CVE-2026-15969 through CVE-2026-15978 > CVE-2026-15969 delivers unauthenticated RCE through /loadloraadapterfromtensors > CVE-2026-15971 triggers a sandbox escape when the optional dumper subsystem is enabled via DUMPERSERVERPORT; inference requests become code execution > CVE-2026-15974 is an SSRF and local file read in the multimodal generation endpoint /v1/chat/completions - unsanitized image_url input leaks internal metadata > CVE-2026-15976 hits when loading model weights from a HuggingFace repository > CVE-2026-15977 exposes API keys and SSL keyfile information through the /server_info endpoint when only --admin-api-key is configured https://news.ojobit.com/story/sglangs-six-unpatched-bugs-open-rce-model-theft-and-credential-leaks-63e3d3

    Post summary

    CERT/CC has publicly disclosed six distinct CVEs in SGLang, detailing the technical impact of each but providing no PoC, exploit code, or patch information.

    1001092
    8 followersView on X
  • CVE@CVEnew
    Disclosure

    CVE-2026-15978 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… https://www.cve.org/CVERecord?id=CVE-2026-15978

    Post summary

    The post discloses that SGLang exposes two endpoints enabling model‑weight exfiltration when API keys are not configured, outlining the vulnerability type and potential impact.

    00010823
    57.9K followersView on X
  • Infoflowcloud@infoflowcloud
    Disclosure

    🚨*CVE* CVE-2026-15978 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… https://www.cve.org/CVERecord?id=CVE-2026-15978 ----- Traducción: CVE-2026-15978 SGL… http://infoflow.cloud`

    Post summary

    The post announces the existence of CVE‑2026‑15978 in SGLang, highlighting a model‑weight exfiltration flaw that occurs when API keys are not configured, and provides a link to the official CVE record.

    0000087
    96 followersView on X
CPE platform detail1 entries

1 of 1 entries

PartVendorProductVersionTarget SWTarget HW
Applmsyssglang---

Explore more

Back to dashboardlmsys CVEssglang vulnerabilities