VAITP

DB-GPT v0.7.5 and v0.8.0 contains directory traversal in python_file_upload (packages/dbgpt-app/src/dbgpt_app/openapi/api_v1/python_upload_api.py:42). A remote attacker can use the validated exploitation path to write files outside the intended workspace or storage boundary.

In camel-ai camel 0.2.91a1, v0.2.91a2 and v0.2.91a3, CodeExecutionToolkit can run model-produced Python code through SubprocessInterpreter without an approval boundary.

In agentscope 1.0.18, 1.0.19, and 1.0.19 when the RealtimeAgent session exposes execute_python_code as an available tool, a remote WebSocket user can prompt the agent to call that tool and run Python code in the service environment. In the validated path, RealtimeAgent._acting forwards the model-produced tool call to Toolkit.call_tool_function, which invokes execute_python_code without an additional approval or isolation boundary on that path.

agent-zero 1.7, 1.8, 1.9, and 1.10 is vulnerable to Directory Traversal in python/helpers/file_browser.py:FileBrowser.__init__. The FileBrowser class initializes with the host root directory as the workspace, allowing the agent to access any file on the system without restriction.

agent-zero 1.7, 1.8, 1.9, and 1.10 is vulnerable to Directory Traversal in python/helpers/file_browser.py:FileBrowser.save_file_b64. The save_file_b64 method accepts user-controlled file paths without normalization or validation, allowing path traversal attacks.

Autobahn Python is a WebSocket and WAMP implementation for Python that supports Twisted and asyncio. Prior to 26.7.1, WebSocket endpoints that accept permessage-deflate and rely on maxMessagePayloadSize enforce that limit against the compressed frame length before inflation but do not recheck the decompressed message size before delivery. A remote unauthenticated client can send a valid compressed frame below the configured wire-size limit that expands beyond the application message limit, causing oversized data to be allocated, joined, validated, and passed to application callbacks. This can create resource-exhaustion pressure, but the advisory does not establish confidentiality or integrity impact. This issue is fixed in version 26.7.1.

SysReptor is a fully customizable pentest reporting platform. Prior to 2026.61, authenticated users of SysReptor Professional can upload image files whose formats cause image processing to invoke Ghostscript, allowing embedded PostScript to operate in the shared temporary directory. An attacker can combine that behavior with a race involving GnuPG configuration files in temporary subdirectories to cause GnuPG to copy attacker-controlled Python code into the application code directory. The injected code executes with the privileges of the SysReptor application process after a worker restart. The Community edition is not affected. Version 2026.58 contains a partial mitigation, and this issue is fully fixed in version 2026.61.

Mnemosyne is a memory layer for artificial intelligence agents. Prior to v3.10.1, the auth check in mnemosyne/core/sync_server.py parsed the JWT's header and payload using base64 decoding, then passed the token to a jwt library call with options that effectively disabled signature verification. The server accepted any well-formed token regardless of the signature, including tokens with alg: none and tokens signed with the wrong key. The fix in v3.10.1 replaces the broken decode with a from-scratch HS256 verifier using only the Python standard library. For users who cannot upgrade immediately, restrict network access to the sync server endpoint to trusted clients only. Firewall, reverse proxy with mTLS, or localhost bind with SSH tunnel are all viable. The vulnerability is not exploitable against an unreachable endpoint.

LMDeploy is a toolkit for compressing, deploying, and serving large language models. Versions 012.1 through 0.12.2 contain a code injection vulnerability in `lmdeploy/pytorch/config.py` line 620 that allows an attacker to execute arbitrary Python code by publishing a malicious HuggingFace model with a crafted `quantization_config.quant_dtype` value. When a user loads the model with lmdeploy, the `quant_dtype` is passed to `eval(f'torch.{quant_dtype}')` without any validation. Version 0.12.3 contains a patch.

LMDeploy is a toolkit for compressing, deploying, and serving large language models. Starting in version 0.9.2 and prior to version 0.16.0, LMDeploy's PyTorch DistServe/PD-disaggregation control plane used `recv_pyobj()` to deserialize messages received through a ZeroMQ PULL socket. PyZMQ implements `recv_pyobj()` using Python pickle deserialization, which can execute arbitrary code while reconstructing an object. The peer address used by the receiver was supplied through the `POST /distserve/p2p_connect` HTTP endpoint. An attacker who could reach an affected DistServe API server could cause the server to connect to an attacker-controlled ZeroMQ endpoint and deserialize a crafted pickle payload. API-key authentication is not enabled unless the operator explicitly configures it. As a result, affected DistServe deployments without API keys allowed unauthenticated remote code execution with the privileges of the LMDeploy serving process. This issue affects the PyTorch backend when PD-disaggregation/DistServe is enabled. Ordinary deployments that do not use the affected disaggregated-serving path do not expose this data flow. The fix was released in LMDeploy 0.16.0. Users who cannot upgrade immediately should prevent untrusted clients from reaching `/distserve/*` endpoints, restrict the DistServe HTTP and ZeroMQ control planes to trusted cluster networks, configure API-key authentication, and block arbitrary outbound ZeroMQ connections from serving nodes. These measures reduce exposure but do not make pickle deserialization safe.