AI Sec Watch

YouTube is expanding its likeness detection technology, a tool that identifies AI-generated deepfakes (videos where AI creates a fake video of someone's face and body), to politicians, government officials, and journalists so they can request removal of unauthorized deepfake content. The tool works similarly to YouTube's Content ID system (which detects copyrighted material), scanning for simulated faces made with AI, and YouTube will evaluate removal requests based on whether the content qualifies as protected speech like parody or political critique.

AI agents are only as effective as the data supporting them, and most companies scaling AI fail not because AI models are weak, but because they lack proper data architecture and governance. The key to success is delivering business context along with data (not just collecting more data), and overcoming 'trust debt' by ensuring data has shared definitions, semantic consistency, and reliable operational context across the many data sources and cloud systems companies use.

OpenAI has released Codex Security, a tool that automatically scans software to find vulnerabilities (security weaknesses that attackers could exploit). In recent testing, it has identified hundreds of critical vulnerabilities across different software programs.

Researchers tested code generated by five different LLM (large language model, AI systems that write text) families across four programming languages and found significant security weaknesses, especially in C and C++ where memory safety issues (bugs that let attackers access or corrupt memory) and hard-coded secrets (passwords or keys written directly in code) were common. The study revealed that LLMs often fail to use modern security features available in newer versions of programming tools and tend to rely on outdated, less secure methods. The researchers conclude that LLMs need to be improved to generate code that follows current security best practices.

FreqTransNet is a new deep learning watermarking model that combines convolutional modules, Transformer structures (neural networks that use self-attention to understand relationships between distant parts of data), and frequency-domain transformations (mathematical techniques that analyze images by breaking them into component frequencies) to embed invisible marks into images more robustly. The model outperforms existing watermarking methods, achieving better visual quality and maintaining over 97% accuracy in extracting watermarks even when images are attacked or modified.

Researchers developed PING (Positive-Incentive Noise Generator), a new method that adds carefully designed noise to protect private data in decentralized learning (where multiple computers train AI models together without sending raw data to a central server) while keeping the learning process efficient. The method uses network connections and lightweight encryption to create correlated noise (noise patterns that work together), and builds on this to create PP-DPIN, an algorithm that combines differential privacy (a mathematical technique for protecting individual data points) and information theory to ensure strong privacy guarantees for at least half the computers involved.

This paper surveys eXplainable AI (XAI, methods for making AI decisions understandable to humans) in visual recognition, which is increasingly important for safety-critical applications like autonomous driving and medical diagnostics. The survey organizes XAI approaches by intent, object, presentation, and methodology, and also examines how interpretability applies to Multimodal Large Language Models (AI systems that process and combine text, images, and other data types).

This research addresses negative transfer, which occurs when an AI model performs worse after trying to apply knowledge from one domain (a labeled dataset) to a different domain (an unlabeled dataset) due to significant differences between them. The study identifies that models relying too heavily on non-causal environmental features (irrelevant details that don't actually cause predictions) creates disagreement across domains, harming performance. The proposed solution, called RED (Reducing Environmental Disagreement), separates each sample into causal features (the truly relevant information) and non-causal environmental features, then reduces the disagreement between domains based on these environmental features.

Adobe has launched a beta version of an AI assistant for Photoshop on the web and mobile apps that uses natural language prompts (instructions written in plain English rather than code) to help users edit images, such as removing objects, changing colors, or adjusting lighting. The company is also expanding its Firefly tool (a media generation and editing platform) with new AI-powered features like generative fill, object removal, and background removal. Paid Photoshop users get unlimited AI generations through April 9, while free users receive 20 generations to start.