The Agentic AI Foundation (AAIF), under the Linux Foundation, is steering efforts to establish a unified framework for AI agents and applications using Google's A2A protocol. With members like OpenAI, Microsoft, and AWS, AAIF aims to foster interoperability and reduce reliance on single-vendor solutions, as AI giants expand capabilities and cut costs to enhance competition in AI ecosystems.

Longbridge has launched a new AI-driven investment platform integrated with ChatGPT and Claude, offering personalized, portfolio-aware insights directly within these popular AI interfaces. This strategic move aims to seamlessly incorporate investment capabilities into existing AI services, avoiding the need for new apps or workflows, and is part of Longbridge's mission to enhance intelligent investing.

Recent AI agent breaches at Hugging Face, OpenAI, and Anthropic reveal critical flaws in access architecture and privilege management, surpassing traditional model risks. Boards must prioritize continuous authorization and identity governance to safeguard sensitive systems and maintain competitive edge in AI-driven landscapes.

AI harnesses are becoming a critical strategic layer, significantly impacting performance, cost, and interoperability across model providers. They function by defining instructions, managing tools, and ensuring security, with business models evolving towards open-source cores and commercial add-ons, highlighting their growing importance in AI workflows.

South Korea launches a $20 billion AI initiative to develop physics- and math-based AI capable of predicting scientific phenomena. The six-year project will involve top scientists and focus on creating new AI architectures and training methods, with support from advanced computing resources.

At ChinaJoy in Shanghai, China's gaming industry emphasizes AI as the linchpin for future growth and global expansion. Baidu AI Cloud leads with partnerships and platforms transforming game development through AI.

The U.S. faces a critical need to modernize its soft power infrastructure, historically built on educational and cultural programs, to maintain global influence. Meanwhile, Gulf states are embedding themselves in global knowledge networks, potentially shifting the power dynamics and challenging U.S. dominance.

Researchers have developed TH-GNN, a groundbreaking temporal graph neural network, to effectively detect shilling attacks in recommender systems, achieving an impressive F1 score of 0.870. This novel approach integrates temporal, structural, and semantic signals, significantly outperforming text-only and graph-only detectors by leveraging cross-modal fusion and modeling temporal burstiness typical of shilling campaigns.

AI systems' rapid deployment is outpacing security measures, creating vulnerabilities as traditional tactics fail to adapt. Organizations must rethink security for the fluid, interconnected AI landscape.

EPAM has partnered with Anthropic, OpenAI, and Google to train specialists for Claude, OpenAI, and Gemini AI models, aiming to create a model-agnostic AI services capability. This initiative reflects the shifting IT landscape where AI blurs traditional boundaries, demanding skills beyond coding to deliver production-ready solutions.