SGLang: Denial of service via decompression amplification in load_audio
Summary
SGLang through version 0.5.21 does not limit the decoded size of compressed audio processed by load_audio. A remote attacker can send small, highly compressed FLAC audio files via audio_url or audio_data that expand into very large PCM buffers. This exhausts worker memory and CPU and can cause a denial of service. No confidentiality or integrity impact is reported.
Affected
- SGLang through and including 0.5.21 (sglang on PyPI)
Fixed in
- Unknown. The vendor has not confirmed a fixed version. Do not assume 0.5.21 or 0.5.22 is fixed.
Workaround
- No vendor workaround documented.
- Restrict access to the SGLang API to trusted networks via an authenticated reverse proxy or VPN.
- Set memory and CPU limits on inference worker processes or containers.
- Block audio input fields at the proxy layer if audio processing is not needed.
Exploitation
- CISA SSVC data lists exploitation as PoC and automatable as yes; VulDB reports technical details known but no available exploit.
- Not listed in CISA KEV. No in-the-wild exploitation reported.
- EPSS below 1 percent.
Our recommendation
- Check whether SGLang instances are reachable from untrusted networks and whether they accept audio input.
- Restrict exposure and place the API behind authentication.
- Set memory and CPU limits on inference workers.
- Monitor worker memory and CPU usage for spikes.
- Watch the vendor issue #43708 and upgrade once a fixed version is confirmed.
Public exploits
PoC-in-GitHub, Metasploit and Exploit-DB list nothing for this CVE.
Research sources
- SGLang GitHub repository github.com
- GitHub issue #43708 github.com
- VulnCheck advisory CVE-2026-109226 www.vulncheck.com
- OSV record CVE-2026-109226 osv.dev
- audio_from_video.py at v0.5.21 github.com
- PSIRT CVE-2026-109226 psirt.com
Researched and written up in English and German by an AI (Claude by Anthropic) based on NVD, CISA, EPSS and the sources listed. The vendor information is authoritative. Researched on 1 Nov 2026, 04:53.
Original description (NVD)
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 audio_url or audio_data inputs that expand into massive PCM buffers, exhausting worker memory and CPU for denial of service.
NVD references (8)
- https://github.com/sgl-project/sglang
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/multimodal/audio_from_video.py#L128
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/multimodal/processors/base_processor.py#…
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/utils/common.py#L1783
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/utils/common.py#L1801-L1803
- https://github.com/sgl-project/sglang/issues/43708
- https://www.vulncheck.com/advisories/sglang-through-0.5.21-audio-decompression-amplification-via-load-audio
- https://github.com/sgl-project/sglang/issues/43708
Sources
The tracker combines official, freely usable data sources – updated continuously:
- NVD (NIST) – CVEs, rating (CVSS), affected products
- CISA KEV – vulnerabilities confirmed by CISA as actively exploited
- CISA Vulnrichment – CISA assessment (SSVC) in the official CVE record
- FIRST EPSS – probability of exploitation
- Microsoft MSRC – Patch Tuesday: vulnerabilities, severity, updates
- PoC-in-GitHub – public proof-of-concept repositories
- Metasploit (Rapid7) – modules of the Metasploit Framework
- Exploit-DB (OffSec) – archive of public exploits (links only)
- Patch days: official vendor pages (linked per vendor) (Vendor patch days)
This product uses data from the NVD API but is not endorsed or certified by the NVD. EPSS data © FIRST.org.
Affected? We can help.
We check whether your systems are affected, install updates and close the gap – continuously with the System Health Checker if you wish.