Coding Agent Usage Simulation

Certifi removes GLOBALTRUST root certificate

HuggingFace transformers vulnerable to remote code execution

Starlette has possible denial-of-service vector when parsing large files in multipart forms

Hydra: hydra.utils.instantiate with untrusted config can lead to code execution

aiohttp: Host-Only Cookies Become Domain Cookies After CookieJar Persistence

vLLM affected by RCE via auto_map dynamic module loading during model initialization

AIOHTTP has CRLF injection through multipart part content type header construction

urllib3 streaming API improperly handles highly compressed data

PyTorch susceptible to local Denial of Service

Transformers is vulnerable to ReDoS attack through its DonutProcessor class

Decompression-bomb safeguards bypassed when following HTTP redirects (streaming API)

Sentencepiece has a a heap overflow issue

cbor2 has a Denial of Service via Uncontrolled Recursion in cbor2.loads

vLLM is vulnerable to Server-Side Request Forgery (SSRF) through `MediaConnector` class

LMCache: 16-bit multimodal hash collision can poison KV cache entries

vLLM: Unauthenticated OOM Denial of Service via Unbounded `n` Parameter in OpenAI API Server

AIOHTTP has late size enforcement for non-file multipart fields causes memory DoS

vLLM's Artifact Pin Decay allows pinned deployments to load unpinned code, weights, and processors

Pillow `BdfFontFile`: `Image.new()` called without `_decompression_bomb_check()` — bomb protection bypass via font loading

vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m

urllib3 does not control redirects in browsers and Node.js

aiohttp: HTTP/1 Pipelined Requests Queue Without Limit

vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections

aiohttp: TLS Server Hostname Override Is Ignored When Reusing HTTPS Connections

Potential Timing Side-Channel Vulnerability in vLLM’s Chunk-Based Prefix Caching

vLLM has RCE In Video Processing

Hugging Face Transformers vulnerable to Regular Expression Denial of Service (ReDoS) in the AdamWeightDecay optimizer

Pillow: WindowsViewer.get_command() OS command injection via unescaped shell path

Pygments has Regular Expression Denial of Service (ReDoS) due to Inefficient Regex for GUID Matching

PyTorch: `torch.load` with `weights_only=True` leads to remote code execution

AIOHTTP vulnerable to brute-force leak of internal static file path components

Denial of service (DoS) via deformation `multipart/form-data` boundary

Hugging Face Transformers is vulnerable to ReDoS through its MarianTokenizer

XGrammar affected by Denial of Service by infinite recursion grammars

vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant serving

AnyIO process-pool workers can block indefinitely on undrained stderr

setuptools has a path traversal vulnerability in PackageIndex.download that leads to Arbitrary File Write

python-multipart: Quadratic-time querystring parsing with semicolon separators causes CPU denial of service

Pillow: `FontFile.compile()`: `Image.new()` called without `_decompression_bomb_check()`

Pillow has a heap buffer overflow with nested list coordinates

Pillow: Out-of-bounds read via attacker-controlled row stride on Pillow's mmap path (McIdas AREA files)

AIOHTTP's C parser (llhttp) accepts null bytes and control characters in response header values - header injection/security bypass

aiohttp: C HTTP Parser Bypasses max_line_size for Fragmented Lines

Internationalized Domain Names in Applications (IDNA): Specially crafted inputs to idna.encode() can bypass CVE-2024-3651 fix

AIOHTTP's unicode processing of header values could cause parsing discrepancies

vLLM vulnerable to DoS via large Chat Completion or Tokenization requests with specially crafted `chat_template_kwargs`

HuggingFace Transformers allows for arbitrary code execution in the `Trainer` class

vLLM: Processing differential in multi-channel audio downmixing enables hidden-input/moderation bypass for audio models