CVE-2025-58446
📋 TL;DR
A denial-of-service vulnerability in xgrammar library versions 0.1.23 allows attackers to cause resource exhaustion by processing large grammars (>100k characters) at very low rates. This affects any system using xgrammar for structured generation, particularly model providers and applications that process user-supplied grammars. The vulnerability is fixed in version 0.1.24.
💻 Affected Systems
- xgrammar library
📦 What is this software?
Xgrammar by Mlc Ai
⚠️ Risk & Real-World Impact
Worst Case
Complete service unavailability due to CPU/memory exhaustion, disrupting model inference services and causing extended downtime.
Likely Case
Degraded performance and intermittent service disruptions when processing maliciously crafted large grammars.
If Mitigated
Minimal impact with proper input validation, rate limiting, and updated library versions.
🎯 Exploit Status
Exploitation requires only the ability to submit large grammar inputs to vulnerable endpoints.
🛠️ Fix & Mitigation
✅ Official Fix
Patch Version: 0.1.24
Vendor Advisory: https://github.com/mlc-ai/xgrammar/security/advisories/GHSA-9q5r-wfvf-rr7f
Restart Required: No
Instructions:
1. Update xgrammar to version 0.1.24 using pip: pip install --upgrade xgrammar==0.1.24
2. Verify the update with: pip show xgrammar
3. Restart any services using xgrammar to load the new version.
🔧 Temporary Workarounds
Input size limitation
allImplement input validation to reject grammars larger than 100k characters
# Python example:
if len(grammar_input) > 100000:
raise ValueError('Grammar too large')
Rate limiting
allImplement rate limiting on grammar processing endpoints
🧯 If You Can't Patch
- Implement strict input validation to reject grammars >100k characters
- Deploy WAF rules to block large grammar payloads at network perimeter
🔍 How to Verify
Check if Vulnerable:
Check xgrammar version: pip show xgrammar | grep Version. If version is 0.1.23, system is vulnerable.Check Version:
pip show xgrammar | grep VersionVerify Fix Applied:
Verify xgrammar version is 0.1.24 or higher: pip show xgrammar | grep Version📡 Detection & Monitoring
Log Indicators:
- Unusually long processing times for grammar operations
- High CPU/memory usage spikes during grammar processing
- Repeated failed grammar processing attempts
Network Indicators:
- Large payloads (>100k) sent to grammar processing endpoints
- High volume of requests to grammar endpoints
SIEM Query:
source="application.logs" AND ("grammar processing" OR "xgrammar") AND (duration>10s OR memory_usage>90%)