Introducing AVISE: A New Framework for Assessing AI System Security

The paper presents AVISE (AI Vulnerability Identification and Security Evaluation), a modular open-source framework designed to identify vulnerabilities and evaluate the security of artificial intelligence systems. With the growing deployment of AI in critical areas, the need for systematic security evaluations is urgent. The authors extend the multi-turn Red Queen attack theory into an Adversarial Language Model (ALM) augmented attack, creating an automated Security Evaluation Test (SET) that includes 25 test cases. This SET demonstrates high accuracy, achieving 92% accuracy, an F1-score of 0.91, and a Matthews correlation coefficient of 0.83, exposing vulnerabilities in nine evaluated language models. AVISE serves as a foundational tool for researchers and industry experts to enhance the rigor and reproducibility of AI security assessments.

Read Full Article

View All For This Day

LLMGenerative AI+2

LLMGenerative AINLPProduct Launch

OpenAI Unveils GPT-5.5: The Most Advanced Model for Complex Tasks

OpenAI has announced the release of GPT-5.5, its most advanced language model to date. The new model is designed to be faster and more capable, specifically optimized for complex tasks including coding, research, and data analysis across various tools. This launch reflects OpenAI's commitment to enhancing the functionalities and performance of its AI models.

AINLP+2

AINLPRegulationEthics

US Government Aims to Prevent Chinese Exploitation of American AI Models

Michael Kratsios, director of the White House Office of Science and Technology Policy, has signed a memo outlining measures to prevent Chinese competitors from misusing outcomes from American AI models. The initiative aims to safeguard the intellectual property and innovations generated by U.S. AI technologies, particularly in the development of competing chatbots.

Bloomberg TechnologyRead →

CybersecurityGenerative AI+2

CybersecurityGenerative AIAI ModelBanking

UK Engages with Anthropic for Mythos Cybersecurity AI Access for Banks

British lenders are in discussions with US-based Anthropic regarding access to its advanced cybersecurity AI model, Mythos. This initiative aims to enhance the security frameworks of financial institutions as they seek expert guidance from organizations currently testing this powerful AI technology. The collaboration highlights the increasing importance of AI in safeguarding against cyber threats in the banking sector.

Financial Times TechRead →

AIDigital Infrastructure+2

AIDigital InfrastructureSustainabilityWorkforce Development

Google Announces First Data Center in Austria's Alps, Creating 100 Jobs

Google has unveiled plans to establish its first data center in Kronstorf, Austria, generating 100 direct jobs. This facility aims to meet the increasing demand for Google’s digital services and AI capabilities, reinforcing Austria's position in innovation. The investment is part of a broader strategy to enhance digital infrastructure across Europe, with a focus on sustainability and community health. Initiatives include a fund to improve local water quality in collaboration with the Upper Austrian Fisheries Association, a green roof equipped with solar panels, and designs for off-site heat recovery. Additionally, Google is partnering with the University of Applied Science Upper Austria to provide training for the local workforce, building on a history of training over 140,000 Austrians for an AI-driven economy.

Google AI BlogRead →

MultimodalReinforcement Learning+2

MultimodalReinforcement LearningTransformersNLP

V-tableR1 Introduces Process-Supervised Multimodal Table Reasoning with Enhanced Reinforcement Learning

Researchers have developed V-tableR1, a novel process-supervised reinforcement learning framework aimed at improving multimodal large language models (MLLMs) by fostering rigorous and verifiable reasoning. Traditional MLLMs often rely on superficial pattern matching for visual reasoning, but V-tableR1 addresses this by utilizing the deterministic grid structure of tables as a testbed for visual domains. The framework incorporates a specialized critic visual language model (VLM) to deliver detailed feedback on visual reasoning, alongside a new reinforcement learning algorithm known as Process-Guided Direct Alignment Policy Optimization (PGPO). This system penalizes visual hallucinations and shortcuts, transitioning multimodal inference from a black-box approach to a logical derivation process. Evaluations indicate that V-tableR1 achieves state-of-the-art accuracy on complex tabular benchmarks, outperforming models significantly larger than itself while also improving upon its supervised fine-tuning baseline.

arXiv AIRead →

AICybersecurity+1

AICybersecurityEthics

White House Alleges Large-Scale Theft of AI Technology by China

Michael Kratsios, a former Trump administration official, has accused Chinese entities of engaging in 'industrial-scale' theft of artificial intelligence technology from American laboratories. This allegation highlights ongoing concerns regarding intellectual property theft and the competitive landscape in the AI sector between the United States and China.