Microsoft’s AI Red Team has released a comprehensive guide addressing failure modes in agentic AI systems. The report outlines the complexities of ensuring reliability and safety in autonomous AI, providing a taxonomy of failure modes, mitigation strategies, and a case study on memory poisoning attacks to enhance system resilience.
Adobe has introduced AI agent features in Photoshop, allowing users to interact with a chatbot for assistance with various tasks. The updated Firefly service now includes new models for enhanced creativity, enabling users to work with static images, videos, and audio while ensuring data privacy during AI training.
At COMPUTEX 2025, Yu-Chuang’s chairman emphasized Taiwan’s pivotal role in the semiconductor industry, dismissing concerns about being hollowed out. The company showcased advanced memory solutions and innovative AI technologies, highlighting their commitment to maintaining a strong presence in the global market amidst geopolitical challenges.
Eduardo Viola, a political scientist, warns that the technological race between global powers will exacerbate disparities, particularly for Latin America. He emphasizes the rapid advancements in artificial intelligence and quantum computing, predicting a widening gap between the region and leading nations like the U.S. and China.
CJ Logistics has partnered with Rainbow Robotics to develop and commercialize AI humanoid robots tailored for the logistics industry. The collaboration aims to create innovative robotic solutions that can autonomously perform tasks, leveraging advanced AI technology for real-time decision-making and enhancing operational efficiency in logistics environments.
The blog post explores the Model Context Protocol (MCP) and Function Calling, two methods for integrating Large Language Models (LLMs) with external systems. It highlights their architectural differences, use cases, and security features, emphasizing MCP’s robustness for complex integrations and Function Calling’s flexibility for dynamic task execution.
Former Google CEO Eric Schmidt has raised alarms about AI’s rapid evolution, warning that machines may soon operate independently of human oversight. He emphasizes the urgent need for regulatory frameworks to manage AI’s growth and the potential risks it poses to society and national security.
Google has launched the Agent Development Kit (ADK), an open-source framework designed for building, managing, and deploying multi-agent systems. Written in Python, ADK emphasizes modularity and flexibility, integrating seamlessly with Google’s AI ecosystem and offering tools for developers to create efficient agent workflows.
Google Cloud is preparing for the agentic AI era, unveiling its new Ironwood TPU, which boasts significant power efficiency and scalability. The company introduced various AI models and software, including an Agent Development Kit and an AI Agent Marketplace, aimed at enhancing the development and management of AI agents.
In a New York appeals court, a man attempted to present his case using an AI-generated avatar, which quickly alarmed the judges. The avatar misled the court, prompting a judge to express frustration over the deception. This incident highlights the growing complexities of AI’s role in the legal system.
