Pypi
7 articles
A LiteLLM Enterprise proxy will leak its master key to a request with no API key. Upgrade to 1.106.5 (GHSA-w4c7-34wj-hqm8)
How to fix GHSA-w4c7-34wj-hqm8: upgrade litellm to 1.106.5, 1.105.6, 1.104.8 or 1.103.10, restart, then rotate LITELLM_MASTER_KEY. An unauthenticated request could make an Enterprise-licensed LiteLLM proxy read its own process environment and send it to an attacker host. CVSS 10.0.
HOL GuardLiteLLM 1.107.0 on PyPI and Docker latest shipped a credential stealer. Pin 1.106.2 and rotate your AI keys
How to fix the LiteLLM 1.107.0 compromise: pin litellm to 1.106.2, rebuild from a clean image, and rotate every key the host could read. All 1.107.0 wheels on PyPI and the Docker latest, main-latest and main-stable tags shipped a credential stealer that also dumps the proxy's stored provider keys.
HOL GuardCVE-2026-76832: Agno PythonTools Path Traversal Escapes base_dir
How to fix CVE-2026-76832: upgrade agno to 2.3.24 or later (current PyPI 2.9.0). PythonTools path traversal can read, write, or run files outside base_dir.
HOL GuardCVE-2026-76850 - LMDeploy Pickle RCE in Disaggregated Serving
How to fix CVE-2026-76850: upgrade InternLM lmdeploy to 0.16.0. Unauthenticated pickle RCE in disaggregated serving. Affects >=0.9.2 and <0.16.0.
HOL GuardCVE-2026-69112: Hugging Face Accelerate Path Traversal Lets Attackers Read Arbitrary Files
Hugging Face Accelerate through 1.14.0 fails to sanitize weight_map entries in sharded checkpoint indexes, allowing arbitrary file reads and denial of service via named pipes. Affects 27M monthly downloads.
Michael KantorCVE-2026-9318: tablib Stored XSS via HTML Export Dataset Title
tablib, a Python tabular data library with over 147 million PyPI downloads, contains a stored cross-site scripting vulnerability in its HTML export functionality. Attackers can embed JavaScript payloads in dataset titles that execute when the exported HTML file is opened in a browser. Fixed in version 3.10.0.
Michael KantorSlopsquatting: When AI Hallucinations Become Supply Chain Attacks
AI coding assistants hallucinate package names 19.7% of the time. Attackers register those names on npm and PyPI before real packages can claim them. Tens of thousands of developers have already installed malicious packages their AI suggested. Here is how the attack works, what the research shows, and how to stop your team from becoming the next victim.
HOLRelated topics
Keep exploring
The engineering desk continues across the ecosystem: security research on shipping agents safely, and plugins that put the standards to work.