Sglang
by Sgl Project
Source repositories
CVEs (39)
| CVE | Vendor / Product | Sev | Risk | CVSS | EPSS | KEV | Published | Description |
|---|---|---|---|---|---|---|---|---|
| CVE-2026-93034 | Cri | 0.64 | 9.8 | 0.00 | Oct 8, 2026 | SGLang contains an arbitrary code execution vulnerability caused by the ZMQ message decoder unconditionally deserializing PickleWrapper payloads via pickle.loads() in _maybe_unwrap_pickle without type allowlisting or authentication; this vulnerability persists via the msgpack… | ||
| CVE-2026-93088 | Cri | 0.64 | 9.8 | 0.01 | Sep 22, 2026 | SGLang's multimodal generation runtime is vulnerable to unauthenticated arbitrary code execution because the disaggregated-diffusion orchestrator's DiffusionServer binds an unauthenticated ZeroMQ ROUTER socket to a network interface and passes the final frame of received… | ||
| CVE-2026-86793 | Cri | 0.64 | 9.8 | 0.01 | Sep 11, 2026 | SGLang allows unauthenticated pickle deserialization through /update_weights_from_tensor 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-2026-7304 | Cri | 0.64 | 9.8 | 0.01 | May 18, 2026 | SGLangs multimodal generation runtime is vulnerable to unauthenticated remote code execution when the --enable-custom-logit-processor option is enabled, as Python objects loaded via dill.loads() will be deserialized without validation. | ||
| CVE-2026-7301 | Cri | 0.64 | 9.8 | 0.00 | May 18, 2026 | SGLangs multimodal generation runtime scheduler's ROUTER socket binds to 0.0.0.0 by default and contains a sink that calls pickle.loads() on incoming messages, enabling RCE when exposed to the internet. | ||
| CVE-2026-14890 | Cri | 0.59 | 9.1 | 0.01 | Jul 16, 2026 | SGLang uses an expert-parallel backup subsystem that exposes a ZeroMQ PULL socket on a routable network interface that does not contain authentication or deserialization safeguards, allowing an attacker to provide a malicious pickle file that results in unauthenticated remote… | ||
| CVE-2026-7302 | Cri | 0.59 | 9.1 | 0.00 | May 18, 2026 | SGLangs multimodal generation runtime is vulnerable to an unauthenticated path traversal vulnerability, allowing an attacker to write arbitrary files anywhere the server process has write access, by including ../ sequences in the upload filename when sent to specific endpoints. | ||
| CVE-2026-5760 | Cri | 0.57 | 9.8 | 0.01 | Apr 20, 2026 | SGLang's reranking endpoint (/v1/rerank) achieves Remote Code Execution (RCE) when a model file containing a malcious tokenizer.chat_template is loaded, as the Jinja2 chat templates are rendered using an unsandboxed jinja2.Environment(). | ||
| CVE-2026-111391 | Hig | 0.53 | 8.1 | 0.00 | Oct 16, 2026 | SGLang 0.5.10 through 0.5.22 contains a code injection vulnerability in _evaluate_filter() that allows remote attackers to execute Python code by escaping the eval() sandbox through re.search __globals__. Attackers can send a crafted filter expression to the unauthenticated… | ||
| CVE-2026-109229 | Hig | 0.49 | 7.5 | 0.01 | Oct 12, 2026 | SGLang through 0.5.21 contains a denial of service vulnerability that disables Pillow decompression bomb protection process-wide by setting Image.MAX_IMAGE_PIXELS to None in NanoNemotronVLImageProcessor. Remote attackers can submit small, highly compressed images with huge… | ||
| CVE-2026-109228 | Hig | 0.49 | 7.5 | 0.00 | Oct 12, 2026 | SGLang through 0.5.21 contains a resource budget bypass vulnerability that allows remote attackers to override GLM-V video processing limits by supplying fps, max_frames, max_tokens_per_frame and max_image_tokens in the video_url object. Attackers can set these values… | ||
| CVE-2026-109227 | Hig | 0.49 | 7.5 | 0.01 | Oct 12, 2026 | SGLang through 0.5.21 contains an unbounded memory allocation vulnerability in preprocess_video in qwen_vl.py that decodes frames at full source resolution before applying max_pixels limits. Remote attackers can submit high-resolution videos to /v1/chat/completions on Qwen-VL… | ||
| CVE-2026-109226 | Hig | 0.49 | 7.5 | 0.01 | Oct 12, 2026 | SGLang through 0.5.21 contains a decompression amplification vulnerability in load_audio that allows remote attackers to exhaust memory by submitting highly compressed audio decoded without duration or sample limits. Attackers can supply small FLAC-encoded silence files through… | ||
| CVE-2026-109225 | Hig | 0.49 | 7.5 | 0.00 | Oct 12, 2026 | SGLang through 0.5.21 contains an unbounded memory allocation vulnerability that allows unauthenticated remote attackers to exhaust memory by uploading oversized files to /v1/audio/transcriptions. Attackers can send concurrent large multipart uploads, which are fully read into… | ||
| CVE-2026-92972 | Hig | 0.49 | 8.6 | 0.00 | Sep 17, 2026 | SGLang through 0.5.19 in prefill/decode disaggregation mode contains an unauthenticated PUT /route endpoint on the prefill bootstrap service that allows attackers to poison the KV transfer routing table. Attackers can supply arbitrary rank_ip and rank_port values to redirect… | ||
| CVE-2026-3989 | Hig | 0.44 | 7.8 | 0.01 | Mar 12, 2026 | SGLangs `replay_request_dump.py` contains an insecure pickle.load() without validation and proper deserialization. An attacker can take advantage of this by providing a malicious .pkl file, which will execute the attackers code on the device running the script. | ||
| CVE-2026-127222 | Hig | 0.42 | 7.5 | 0.01 | Nov 18, 2026 | SGLang through 0.5.23, fixed in commit 7b5fb39, contains a missing authorization vulnerability in the POST /load_lora_adapter_from_tensors HTTP endpoint, which lacks the @auth_level(AuthLevel.ADMIN_OPTIONAL) decorator applied to its sibling admin endpoints, allowing… | ||
| CVE-2026-112270 | Med | 0.42 | 6.5 | 0.00 | Oct 19, 2026 | SGLang through 0.5.22 with the Mooncake disaggregation backend contains an uncontrolled memory allocation vulnerability in handle_staging_rsp() that lets unauthenticated network-adjacent attackers exhaust Prefill memory. Attackers can obtain the rank address from the /route… | ||
| CVE-2026-112269 | Med | 0.42 | 6.5 | 0.00 | Oct 19, 2026 | SGLang through 0.5.22 with the Mooncake disaggregation backend accepts unauthenticated decode registration messages on the Prefill rank ZMQ PULL socket, allowing denial of service. Network-adjacent attackers can send many registrations with unique mooncake_session_id values and… | ||
| CVE-2026-112267 | Med | 0.42 | 6.5 | 0.00 | Oct 19, 2026 | SGLang through 0.5.22 in prefill/decode disaggregation mode contains a server-side request forgery vulnerability that allows unauthenticated attackers to trigger outbound requests via client-supplied bootstrap_host and bootstrap_port fields. Attackers sending crafted requests to… |
- risk 0.64cvss 9.8epss 0.00
SGLang contains an arbitrary code execution vulnerability caused by the ZMQ message decoder unconditionally deserializing PickleWrapper payloads via pickle.loads() in _maybe_unwrap_pickle without type allowlisting or authentication; this vulnerability persists via the msgpack…
- risk 0.64cvss 9.8epss 0.01
SGLang's multimodal generation runtime is vulnerable to unauthenticated arbitrary code execution because the disaggregated-diffusion orchestrator's DiffusionServer binds an unauthenticated ZeroMQ ROUTER socket to a network interface and passes the final frame of received…
- risk 0.64cvss 9.8epss 0.01
SGLang allows unauthenticated pickle deserialization through /update_weights_from_tensor 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.
- risk 0.64cvss 9.8epss 0.01
SGLangs multimodal generation runtime is vulnerable to unauthenticated remote code execution when the --enable-custom-logit-processor option is enabled, as Python objects loaded via dill.loads() will be deserialized without validation.
- risk 0.64cvss 9.8epss 0.00
SGLangs multimodal generation runtime scheduler's ROUTER socket binds to 0.0.0.0 by default and contains a sink that calls pickle.loads() on incoming messages, enabling RCE when exposed to the internet.
- risk 0.59cvss 9.1epss 0.01
SGLang uses an expert-parallel backup subsystem that exposes a ZeroMQ PULL socket on a routable network interface that does not contain authentication or deserialization safeguards, allowing an attacker to provide a malicious pickle file that results in unauthenticated remote…
- risk 0.59cvss 9.1epss 0.00
SGLangs multimodal generation runtime is vulnerable to an unauthenticated path traversal vulnerability, allowing an attacker to write arbitrary files anywhere the server process has write access, by including ../ sequences in the upload filename when sent to specific endpoints.
- risk 0.57cvss 9.8epss 0.01
SGLang's reranking endpoint (/v1/rerank) achieves Remote Code Execution (RCE) when a model file containing a malcious tokenizer.chat_template is loaded, as the Jinja2 chat templates are rendered using an unsandboxed jinja2.Environment().
- risk 0.53cvss 8.1epss 0.00
SGLang 0.5.10 through 0.5.22 contains a code injection vulnerability in _evaluate_filter() that allows remote attackers to execute Python code by escaping the eval() sandbox through re.search __globals__. Attackers can send a crafted filter expression to the unauthenticated…
- risk 0.49cvss 7.5epss 0.01
SGLang through 0.5.21 contains a denial of service vulnerability that disables Pillow decompression bomb protection process-wide by setting Image.MAX_IMAGE_PIXELS to None in NanoNemotronVLImageProcessor. Remote attackers can submit small, highly compressed images with huge…
- risk 0.49cvss 7.5epss 0.00
SGLang through 0.5.21 contains a resource budget bypass vulnerability that allows remote attackers to override GLM-V video processing limits by supplying fps, max_frames, max_tokens_per_frame and max_image_tokens in the video_url object. Attackers can set these values…
- risk 0.49cvss 7.5epss 0.01
SGLang through 0.5.21 contains an unbounded memory allocation vulnerability in preprocess_video in qwen_vl.py that decodes frames at full source resolution before applying max_pixels limits. Remote attackers can submit high-resolution videos to /v1/chat/completions on Qwen-VL…
- risk 0.49cvss 7.5epss 0.01
SGLang through 0.5.21 contains a decompression amplification vulnerability in load_audio that allows remote attackers to exhaust memory by submitting highly compressed audio decoded without duration or sample limits. Attackers can supply small FLAC-encoded silence files through…
- risk 0.49cvss 7.5epss 0.00
SGLang through 0.5.21 contains an unbounded memory allocation vulnerability that allows unauthenticated remote attackers to exhaust memory by uploading oversized files to /v1/audio/transcriptions. Attackers can send concurrent large multipart uploads, which are fully read into…
- risk 0.49cvss 8.6epss 0.00
SGLang through 0.5.19 in prefill/decode disaggregation mode contains an unauthenticated PUT /route endpoint on the prefill bootstrap service that allows attackers to poison the KV transfer routing table. Attackers can supply arbitrary rank_ip and rank_port values to redirect…
- risk 0.44cvss 7.8epss 0.01
SGLangs `replay_request_dump.py` contains an insecure pickle.load() without validation and proper deserialization. An attacker can take advantage of this by providing a malicious .pkl file, which will execute the attackers code on the device running the script.
- risk 0.42cvss 7.5epss 0.01
SGLang through 0.5.23, fixed in commit 7b5fb39, contains a missing authorization vulnerability in the POST /load_lora_adapter_from_tensors HTTP endpoint, which lacks the @auth_level(AuthLevel.ADMIN_OPTIONAL) decorator applied to its sibling admin endpoints, allowing…
- risk 0.42cvss 6.5epss 0.00
SGLang through 0.5.22 with the Mooncake disaggregation backend contains an uncontrolled memory allocation vulnerability in handle_staging_rsp() that lets unauthenticated network-adjacent attackers exhaust Prefill memory. Attackers can obtain the rank address from the /route…
- risk 0.42cvss 6.5epss 0.00
SGLang through 0.5.22 with the Mooncake disaggregation backend accepts unauthenticated decode registration messages on the Prefill rank ZMQ PULL socket, allowing denial of service. Network-adjacent attackers can send many registrations with unique mooncake_session_id values and…
- risk 0.42cvss 6.5epss 0.00
SGLang through 0.5.22 in prefill/decode disaggregation mode contains a server-side request forgery vulnerability that allows unauthenticated attackers to trigger outbound requests via client-supplied bootstrap_host and bootstrap_port fields. Attackers sending crafted requests to…
Page 1 of 2