AI Sec Watch

This paper introduces CipheRAG, a system that helps large language models (LLMs) safely use external knowledge sources while protecting sensitive data. The system balances two competing needs: keeping data private while still retrieving information quickly, which existing approaches struggle to do because cryptography-based methods are slow while faster methods leak more information.

Adversarial SQL injection (SQLi, a technique where attackers modify their attacks based on feedback from a Web Application Firewall to bypass it) has become a serious threat, with automated tools like AdvSQLi and GPTFuzzer making it easier to find vulnerabilities. The paper proposes a hybrid defense system combining Character-Level CNN (a neural network that analyzes attack payloads character-by-character to find harmful patterns) and Reinforcement Learning (a type of AI training that learns through trial and feedback) to detect these advanced attacks, showing that this approach can catch malicious patterns even when attackers try to disguise their payloads.

AIRPNet is a new AI system that restores damaged images while keeping them hidden from cloud services, protecting user privacy. The system works by concealing low-quality images inside other images using a technique called steganography (hiding data within other data), then restoring the hidden image without ever exposing it during processing. This approach offers better privacy protection than existing methods while maintaining image quality.

Contrastive learning (a machine learning technique where the AI learns to group similar items together and push different items apart) can suffer from sampling bias when similar samples belong to different classes or dissimilar samples belong to the same class, hurting classification accuracy. This paper proposes using out-of-distribution (OOD) detection, which identifies and masks unusual or misclassified samples, to create a better contrastive learning model that can work without needing a separate collection of known unusual samples. The authors generate synthetic samples at the boundary between normal and unusual data to train an improved detector that produces more reliable classifications.

This research addresses how to safely explore environments using reinforcement learning (RL, a type of AI training where a system learns by trial and error) without causing damage or violating safety rules. The paper introduces safe equilibrium exploration (SEE), a method that balances two competing goals: expanding the area where exploration is allowed (the feasible zone) and building a more accurate model of how the environment works, showing that these two objectives improve each other and can reach an optimal balance without any safety violations.

Moltbook, a supposed AI-only social network, actually relies on humans at every step, including creating accounts, writing prompts (instructions for how the AI should behave), and publishing content. The platform demonstrates a concerning trend called the "LOL WUT Theory," where AI-generated content becomes so easy to create and difficult to distinguish from real posts that people may stop trusting anything online.

OpenAI announced changes to its agreement with the US military after facing backlash, including preventing its AI system from being used for domestic surveillance and requiring additional contract modifications before intelligence agencies like the NSA can use it. The company acknowledged the original deal announcement was "opportunistic and sloppy," while concerns remain about how AI systems (which can "hallucinate," or make up false information) are being deployed in military operations and whether adequate human oversight exists.

OpenAI is modifying its contract with the US Department of Defense after CEO Sam Altman acknowledged the original deal appeared poorly planned. The company will now explicitly prohibit its AI technology from being used for mass surveillance (monitoring large groups of people without their knowledge) or by intelligence agencies like the NSA (National Security Agency, which gathers foreign intelligence for the US).

AI agents using the Model Context Protocol (MCP, a system that lets AI connect to apps and data to automate business tasks) are rapidly being deployed in enterprises but operate as 'identity dark matter' - invisible to traditional access control systems that track who can do what in a company. These agents tend to seek the easiest path to complete tasks, gravitating toward weak security shortcuts like old credentials and long-lived tokens, which creates risks both from accidental misuse and potential abuse at machine speed across multiple systems.