SGLang-0.5.9-RCE

SGLang SSTI to RCE PoC — 샌드박스가 적용되지 않은 Jinja2 채팅 템플릿 렌더링 CVE-2026-5760 SGLang의 reranking 엔드포인트에서 서버 측 템플릿 주입SSTI을 통한 원격 코드 실행. 악성 GGUF 모델 파일에 조작된 tokenizer.chattemplate이 포함되어 있으며, 해당 모델이 SGLang에 로드되고 /v1/rerank 엔드포인트가 호출되면 임의 코드 실행이 이루어집니다. 기능 SGLang의 샌드박스가 적용되지 않은 Jinja2 템플릿 렌더링을 악용하는 Critical 심...

CVE-2026-86793

SGLang allows unauthenticated pickle deserialization through /updateweightsfromtensor when no auth keys are configured, and the SafeUnpickler policy can be bypassed because builtins.import and builtins.getattr are resolvable, enabling code execution via pickle REDUCE...

EUVD-2026-76073

SGLang allows unauthenticated pickle deserialization through /updateweightsfromtensor when no auth keys are configured, and the SafeUnpickler policy can be bypassed because builtins.import and builtins.getattr are resolvable, enabling code execution via pickle REDUCE...

CVE-2026-86793

SGLang is vulnerable to remote code execution through its /update_weights_from_tensor endpoint. When no authentication keys are configured, an unauthenticated attacker can send a crafted pickle payload. The application's SafeUnpickler policy is bypassable because builtins.import and builtins.geta...

PT-2026-89864

🚨 AI INFRA CRISIS: SGLang just got hit with an unauthenticated RCE CVE-2026-86793 — the 4th critical CVE in AI infrastructure in just 18 days. No vendor patch yet. AI inference servers are now primary attack surface. Time to audit your stack. AI cybersecurity infosec...

CVE-2026-7669-PoC

CVE-2026-7669 SGLang은 trustremotecode=False를 조용히 True로 덮어쓰고, transformers v5가 TokenizersBackend 객체를 반환할 때마다 AutoTokenizer.frompretrained를 다시 호출합니다. 사용자 정의 tokenizerclass와 tokenizer.py를 가리키는 automap을 가진 모델은 운영자가 False를 전달했더라도 SGLang 프로세스 내부에서 임의 코드 실행에 도달합니다. 어떤 로그 레벨에서도 로그 줄이 출력되지 않습니다. 프로젝트:...

VulnCheck KEV: CVE-2025-10164

A security flaw has been discovered in lmsys sglang 0.4.6. Affected by this vulnerability is the function main of the file /updateweightsfromtensor. The manipulation of the argument serializednamedtensors results in deserialization. The attack can be launched remotely. The exploit has been releas...

SGLang is Vulnerable to DoS via the data_hash Function

A vulnerability was determined in sgl-project SGLang up to 0.5.11. Affected by this vulnerability is the function datahash of the component Cache Handler. This manipulation causes denial of service. The attack is restricted to local execution. A high degree of complexity is needed for the attack...

PYSEC-2026-3695 SGLang is Vulnerable to DoS via the data_hash Function

A vulnerability was determined in sgl-project SGLang up to 0.5.11. Affected by this vulnerability is the function datahash of the component Cache Handler. This manipulation causes denial of service. The attack is restricted to local execution. A high degree of complexity is needed for the attack...

arbor-ai (>=0.1.5, <=0.1.14), coreason-runtime (>=0.1.0, <=0.11.1) +13 more potentially affected by CVE-2026-15977 via sglang (>=0.1.26, <=0.5.18)

sglang PYPI version =0.1.26, =0.1.5, =0.1.0, =1.1.0rc1, =2.0.0b40, =0.4.0, =0.0.1, =0.1.0, =0.1.0, =0.0.1.post1, =0.1.1, =0.0.0, =0.8.0, =0.10.8 Source cves: CVE-2026-15977 Source advisory: SNYK:PYTHON-SGLANG-19432281...

Information Exposure

Overview sglang is a SGLang is a fast serving framework for large language models and vision language models. Affected versions of this package are vulnerable to Information Exposure via the /serverinfo endpoint when only the --admin-api-key is configured. An attacker can obtain sensitive API key...

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 distributed weight broadcasting using NCCL and then triggering data transfer, attackers can exfiltrate all model weights...

CVE-2026-15969

SGLang contains an unauthenticated RCE in /loadloraadapterfromtensors via bypass of SafeUnpickler’s incomplete denylist, allowing arbitrary command execution through crafted base64-encoded pickle payloads...

CVE-2026-15976

SGLang contains a RCE vulnerability when attempting to load model weights from a HuggingFace repository, specifically within the /updateweightsfromdisk, where torch.load..., weightsonly=False fallback enables pickle deserialization of .bin files...

CVE-2026-15971

SGLang contains an RCE vulnerability when the optional dumper subsystem is enabled, allowing for a sandbox escape when DUMPERSERVERPORT is set, enabling code execution on inference requests...

CVE-2026-15974

SGLang contains an SSRF and local file read in the multimodal generation endpoint /v1/chat/completions due to unsanitized imageurl, allowing access to internal metadata, secrets, and services...

CVE-2026-15969

SGLang (CVE-2026-15969) exposes an unauthenticated RCE via /load_lora_adapter_from_tensors caused by bypassing SafeUnpickler’s incomplete denylist, enabling arbitrary command execution through crafted base64-encoded pickle payloads. Multiple sources confirm the vulnerability and its impact on SGL...

CVE-2026-15969 CVE-2026-15969

SGLang contains an unauthenticated RCE in /loadloraadapterfromtensors via bypass of SafeUnpickler’s incomplete denylist, allowing arbitrary command execution through crafted base64-encoded pickle payloads...

CVE-2026-15978

SGLang (vaguely referenced in CVE-2026-15978) contains a model weight exfiltration vulnerability that is exploitable when no API keys are configured. The vulnerability arises because SGLang exposes two endpoints that can trigger distributed weight broadcasting via NCCL and then data transfer, all...

CVE-2026-15978 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 distributed weight broadcasting using NCCL and then triggering data transfer, attackers can exfiltrate all model weights...