AI Sec Watch

A new benchmark called the Remote Labor Index (RLI) measures whether AI systems can automate real computer work tasks across different professions, showing that current AI agents can only fully automate 2.5% of projects despite improving over time. Additionally, over 50,000 people, including top scientists and Nobel laureates, signed an open letter calling for a moratorium (temporary ban) on developing superintelligence (a hypothetical AI system far more capable than humans) until it can be proven safe and controllable.

CVE-2025-12058 is a vulnerability in Keras (a machine learning library) where the load_model method can be tricked into reading files from a computer's local storage or making network requests to external servers, even when the safe_mode=True security flag is enabled. The problem occurs because the StringLookup layer (a component that converts text into numbers) accepts file paths during model loading, and an attacker can craft a malicious .keras file (a model storage format) to exploit this weakness.

Anthropic added network request capabilities to Claude's Code Interpreter, which creates a security risk for data exfiltration (unauthorized stealing of sensitive information). An attacker, either controlling the AI model or using indirect prompt injection (hidden malicious instructions in a document the AI processes), could abuse Anthropic's own APIs to steal data that a user has access to, rather than using typical methods like hidden links.

This article reviews prompt engineering (the practice of designing inputs like questions or instructions to guide AI systems toward better responses) and analyzes its strengths, weaknesses, opportunities, and threats using a SWOT framework. The review covers how prompt engineering can improve interactions with large language models (advanced AI systems trained on vast amounts of text) across industries like healthcare and education, while also identifying challenges around maintaining accuracy and efficiency.

This paper presents RINNs (reparameterizable integral neural networks), a new type of AI model designed to run efficiently on mobile devices with limited computing power. The key innovation is a reparameterization strategy that converts the complex mathematical structure used during training into a simpler feed-forward structure (a straightforward sequence of processing steps) at inference time, allowing these models to achieve high accuracy (79.1%) while running very fast (0.87 milliseconds) on mobile hardware.

A SQL injection vulnerability (a type of attack where an attacker inserts malicious SQL code into an application) exists in LangGraph's SQLite storage system, specifically in version 2.0.10 of langgraph-checkpoint-sqlite. The vulnerability happens because the code directly combines user input with SQL commands instead of safely separating them, allowing attackers to steal sensitive data like passwords and API keys, and bypass security protections.

LlamaIndex v0.14.6 is a software update released on October 26, 2025, that fixes various bugs across multiple components including support for parallel tool calls, metadata handling, embedding format compatibility, and SQL injection vulnerabilities (using parameterized queries instead of raw SQL string concatenation). The release also adds new features like async support for retrievers and integrations with new services like Helicone.

FastGPT, an AI Agent building platform, had a vulnerability in its workflow file reading node where network links were not properly verified, creating a risk of SSRF attacks (server-side request forgery, where an attacker tricks the server into making unwanted requests to other systems). The vulnerability affected versions before 4.11.1.

Hugging Face Smolagents version 1.20.0 has an XPath injection vulnerability (a security flaw where attackers can inject malicious code into XPath queries, which are used to search and navigate document structures) in its web browser function. The vulnerability exists because user input is directly inserted into XPath queries without being cleaned, allowing attackers to bypass search filters, access unintended data, and disrupt automated web tasks.