VAITP

PraisonAI (pip package praisonaiagents) before 1.6.78 contains an unsafe dynamic module loading vulnerability in AgentFlow._resolve_pydantic_class (src/praisonai-agents/praisonaiagents/workflows/workflows.py). When a workflow step uses a string output_pydantic reference, the framework locates and imports a sibling tools.py from the workflow file's directory via importlib exec_module without sandboxing, ignoring the PRAISONAI_ALLOW_*_TOOLS environment variables. An attacker who controls a workflow file and its sibling tools.py can execute arbitrary Python code with the workflow runner's privileges when the workflow is executed via WorkflowManager or after load_yaml.

Langroid is a framework for building large-language-model-powered applications. Versions prior to 0.65.2 are vulnerable to a critical Sandbox Escape leading to Remote Code Execution (RCE) in its `TableChatAgent` and `VectorStore` capabilities. When these agents evaluate LLM-generated tool messages with `full_eval=True`, they attempt to sandbox the execution by explicitly setting `locals` to an empty dictionary `{}` inside Python's `eval()` function. However, this relies on an incomplete understanding of Python's execution model. Because `__builtins__` is not explicitly scrubbed from the `globals` dictionary mapping, Python implicitly injects all built-ins during execution, granting full access to functions like `__import__('os').system()`. Since `TableChatAgent.pandas_eval()` executes external LLM outputs natively, this bypass permits any attacker providing prompt payload to achieve unauthenticated RCE on the host system. Version 0.65.2 patches the issue.

Python Liquid is a Python engine for the Liquid template language. Prior to 2.2.1, given a malformed {% case %} tag without an associated {% when %} or {% else %} block and no terminating {% endcase %} tag, Python Liquid hangs in an infinite loop at parse time because liquid.TokenStream.eof did not give the EOF token matching kind and value fields, allowing malicious template authors to craft templates for a denial of service attack. This issue is fixed in version 2.2.1.

Pipecat is an open-source Python framework for building real-time voice and multimodal conversational agents. Prior to 1.4.0, the pipecat development runner registers a /ws WebSocket endpoint for telephony testing that accepts connections without authentication, reads an attacker-supplied callSid from a Twilio stream-start handshake in src/pipecat/runner/utils.py, and passes it to TwilioFrameSerializer so the server can issue an authenticated Twilio REST API hang-up request with the server operator's credentials; equivalent unauthenticated call-control sinks exist for Telnyx and Plivo. This issue is fixed in version 1.4.0.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. Prior to 0.10.0, get_event_call delivered execute:python and execute:tool Socket.IO events to a client-supplied session_id after checking only that the session was connected, allowing authenticated users who learned another socket ID through ydoc:document:join to run code interpreter Python or tools in that user session. This issue is fixed in version 0.10.0.

Open WebUI is an extensible, feature-rich, and user-friendly self-hosted AI platform. Prior to 0.10.0, Open WebUI runs client-side Python with Pyodide in a same-origin web worker, allowing stored chat payloads that use pyodide.http.pyfetch or the js module fetch and XMLHttpRequest APIs to issue authenticated same-origin requests when a victim clicks Run, which can reach admin-only endpoints and execute server-side code through configured tools. This issue is fixed in version 0.10.0.

Stanza is a Stanford NLP Python library for tokenization, sentence segmentation, NER, and parsing of many human languages. Prior to 1.12.2, Stanza model loaders such as stanza.models.common.pretrain.Pretrain.load() attempt torch.load(..., weights_only=True) but fall back to torch.load(..., weights_only=False) on attacker-controllable pickle.UnpicklingError, allowing a malicious .pt pretrain or model file to execute arbitrary pickle code when a Stanza NLP pipeline loads it. This issue is fixed in version 1.12.2.

RestrictedPython is a tool that helps to define a subset of the Python language which allows to provide a program input into a trusted environment. Prior to 8.3, check_function_argument_names() rejected protected guard hook names for regular, variadic, and keyword-only arguments but omitted positional-only arguments, allowing __getattr__, _getitem_, _write_, or _print_ to be shadowed by a local parameter and bypass the embedding application's access policy. This issue is fixed in version 8.3.

py7zr is a Python-based library and utility to support 7zip archive compression, decompression, encryption and decryption. Prior to 1.1.3, PackInfo._read() in archiveinfo.py used an O(n^2) cumulative sum pattern for attacker-controlled numstreams values parsed from archive headers, allowing a crafted .7z archive to cause excessive CPU consumption during SevenZipFile.init() before extraction. This issue is fixed in version 1.1.3.

py7zr is a Python-based library and utility to support 7zip archive compression, decompression, encryption and decryption. Prior to 1.1.3, py7zr's Worker.decompress() extracted archive entries without tracking total decompressed size, allowing a crafted .7z file such as a 15.6 KB archive that expands to 100 MB to exhaust disk or memory before extraction completes. This issue is fixed in version 1.1.3.