AI Sec Watch

A path traversal vulnerability (CWE-22, where an attacker manipulates file paths to access files outside their intended directory) was discovered in Visual Studio Code's CoPilot Chat Extension that allows an authorized attacker to bypass a security feature on their local computer. The vulnerability is tracked as CVE-2025-62449 and was reported by Microsoft Corporation.

CVE-2025-62222 is a command injection vulnerability (where an attacker tricks software into running unintended commands) in the Visual Studio Code CoPilot Chat Extension that allows an unauthorized attacker to execute code over a network. The vulnerability stems from improper neutralization of special elements in commands and inadequate input validation (checking that data is safe before using it).

This research addresses the problem of incomplete knowledge graphs (databases of connected facts about entities) by proposing a new model called TEDD that predicts missing relationships between entities. The model combines both structural information from the graph and text information, and uses a specialized transformer technique (BERT, a language processing model) to reduce computational costs and handle entities that change over time in dynamic knowledge graphs.

This research paper argues that the real problem with machine learning classifiers isn't that robustness (resistance to adversarial attacks, where small malicious changes trick the AI) and accuracy are fundamentally opposed, but rather that continuous functions (smooth mathematical functions without jumps or breaks) cannot achieve both properties simultaneously. The authors propose that effective robust and accurate classifiers should use discontinuous functions (functions with breaks or sudden changes) instead, and show that understanding this continuity property is crucial for building, analyzing, and testing modern machine learning models.

Milvus, an open-source vector database (a specialized database that stores and searches data based on similarity patterns, used in AI applications), has a critical vulnerability in older versions that allows attackers to skip authentication and gain full admin control over the database without needing a password. This means attackers could read, change, or delete any data and perform administrative tasks like managing databases.

This release notes document describes version updates across multiple llama-index (a framework for building AI applications with language models) components, including fixes for bugs like a ReActOutputParser (a tool that interprets AI agent outputs) getting stuck, improved support for multiple AI model providers like OpenAI and Google Gemini, and updates to various integrations with external services. The updates span from core functionality fixes to documentation improvements and SDK compatibility updates across dozens of sub-packages.

Langfuse, an open source platform for managing large language models, had a vulnerability in versions 2.70.0 through 2.95.10 and 3.x through 3.124.0 where the server didn't properly check which organization a user belonged to, allowing any authenticated user to see names and email addresses of members in other organizations if they knew the target organization's ID. The vulnerability required the attacker to have a valid account on the same Langfuse instance and knowledge of the target organization's ID, and no customer data like traces, prompts, or evaluations were exposed.

A WordPress plugin called Tag, Category, and Taxonomy Manager – AI Autotagger with OpenAI has a SQL injection vulnerability (a security flaw where attackers can insert harmful database commands into the plugin's code) in versions up to 3.40.0. Attackers with Editor-level access or higher can exploit the 'post_types' parameter to extract sensitive information from the website's database because the plugin doesn't properly clean up user input before using it in database queries.

ATLAS Data v5.1.0 is an updated framework that documents security threats and defenses related to AI systems, now containing 16 tactics, 84 techniques, and 32 mitigations. The update adds new attack methods targeting AI, such as prompt injection (tricking an AI by hiding instructions in its input), deepfake generation, and data theft from AI services, along with new defensive measures like human oversight of AI agent actions and restricted permissions for AI tools. It also includes 42 real-world case studies showing how these attacks and defenses apply in practice.