Security context
What an agent needs to avoid regressing past fixes and find the next vuln in this repo.
mlc-ai/mlc-llm
main @ 2008fe8
4
Fixes
0
CVEs
HIGH
Peak severity
33.3%
Coverage
Highlights
Denial of Service: 4 prior fixes. Scrutinize any change in this area.
cpp/serve/engine.cc: most-fixed (2 issues). Treat as high-risk during review.
1 high-severity fix in this history; regressions here are high-impact.
Recurring patterns
The bug types that recur here, drawn from past fixes, not open vulnerabilities.
Denial of Service: Resource leaks in the speculative decoding engine's draft token workspace manager fail to release slots upon request completion, leading to eventual capacity exhaustion and engine-wide denial of service.
Denial of Service: A segmentation fault occurs when the engine attempts to process tokenizers missing their mandatory tokenizer.json file, allowing malformed models to crash the service.
Denial of Service: An integer overflow can trigger during memory footprint estimation when sequence length calculations with sliding windows enabled overflow native integer boundaries.