Microsoft

Microsoft Copilot reveals secret input that allowed it to be hacked

Secret parameter allowed hackers to steal passwords when a target clicked on a link.

Microsoft expands Zero Trust strategy with new AI security tools and guidance

Microsoft has enhanced its Zero Trust for AI strategy by introducing new tools and guidance aimed at securing AI agents and DevSecOps environments. This update focuses on strengthening security measures in AI development and deployment.

Why it matters: As AI systems become more integrated into critical operations, securing AI agents and development pipelines is essential to prevent vulnerabilities. Microsoft's expanded Zero Trust approach addresses these risks by providing targeted security solutions for AI and DevSecOps.

Orchard: Open-Source Framework for Scalable AI Agents

Orchard is an open-source framework designed to help researchers train and evaluate AI agents across various tasks. It simplifies development and supports strong performance even with smaller models by reusing infrastructure.

Why it matters: By reducing complexity and enabling infrastructure reuse, Orchard facilitates scalable AI agent research and accelerates experimentation. This can help advance agentic AI capabilities more efficiently within the research community.

Echoverse trains AI agents in evolving environments for complex workflows

Echoverse addresses the challenge AI agents face with multi-step workflows by training them in realistic, evolving environments. This approach helps agents improve as tasks and environments change, rather than relying on static training tasks.

Why it matters: Training AI agents in dynamic environments can enhance their ability to handle complex, real-world tasks like email management and customer support. This method could lead to more adaptable and effective AI agents in practical applications.

Microsoft 365 Copilot Agent Playbook Livestream Series Launches

Microsoft has introduced a practical livestream series called the Copilot Agent's Playbook to help developers build better declarative agents for Microsoft 365 Copilot. These agents integrate organizational knowledge and workflows directly into users' work processes.

Why it matters: As Microsoft 365 Copilot agent capabilities expand, developers need clear guidance to effectively create agents that enhance productivity. This series aims to provide that practical support, facilitating smoother adoption and innovation within enterprise workflows.

Final article in series on improving AI coding agent evaluations

The eighth article in Microsoft's series on Agent Experience (AX) discusses how to effectively evaluate AI coding agents and improve their integration with technology. It covers controlling the agent stack, measuring extension impact, and iterating for better results.

Why it matters: Understanding how to properly evaluate AI coding agents helps developers ensure these tools work correctly and efficiently with their systems. This leads to more reliable AI-assisted coding workflows.

Let’s Learn GitHub Copilot App – Free Virtual Training Event

Join us for a free online event series kicking off July 16 to learn how to get started with the GitHub Copilot App! The post Let’s Learn GitHub...

Flint: AI-powered visualization language for expressive charts

Flint is an open-source visualization language that enables AI agents to generate expressive charts from compact, human-editable specifications. It offers a balance between ease of writing and visual quality.

Why it matters: By leveraging AI to enhance chart expressiveness without sacrificing simplicity, Flint can improve data visualization workflows. This approach helps users create more insightful visuals with less effort.

The hidden variables in your agent eval

This is the seventh article in a series about Agent Experience (AX): the practice of making AI coding agents work correctly with your technology. The series covers what you can and can’t control in the agent stack, how to measure whether your extensi...

Model upgrades can increase token use and reduce output quality

A comparison of Claude Sonnet 4.6 and Claude Sonnet 5 models revealed that the newer model used 12 times more tokens for the same tasks while delivering worse results. This was observed across 150 agent tasks in 15 scenarios using GitHub Copilot.

Why it matters: This highlights that newer AI model versions may not always improve efficiency or output quality despite better benchmarks and lower pricing. Developers should carefully evaluate model upgrades before switching to avoid unexpected performance regressions.

What AI benchmarks are not telling you

This is the sixth article in a series about Agent Experience (AX): the practice of making AI coding agents work correctly with your technology. The series covers what you can and can’t control in the agent stack, how to measure whether your extension...

SkillOpt: Training AI Agent Skills Without Changing Model Weights

SkillOpt introduces a method to treat AI agent skills as trainable parameters, enabling skill improvement through training rather than manual editing. This approach enhances agent behavior reliability without altering the underlying model weights.

Why it matters: By converting skill editing into a training process, SkillOpt addresses the common failure point of manual skill modification in AI agents. This can lead to more consistent and dependable agent performance in various applications.

Memora Enhances AI Agents' Memory with Scalable Context Retrieval

Memora introduces a scalable memory system for AI agents that separates storage from retrieval, improving efficiency in handling long and complex tasks. This approach addresses the challenge of AI agents' limited ability to remember past conversations.

Why it matters: By enabling more efficient context management, Memora can help AI agents perform better in extended interactions and complex tasks. This advancement could lead to more coherent and context-aware AI applications.

Rethinking cloud operations with agentic observability

Cloud operations are entering a new era as AI-driven and autonomous agents become a larger part of modern software systems. As software becomes increasingly agentic, the challenge is no longer just managing greater scale and complexity. Operators must also...

Unlocking a New Way to Build Enterprise Data Apps with Microsoft Fabric

We believe enterprise software should move at the speed of ideas. Today, we’re collaborating with Microsoft to make it dramatically easier for enterprises to build AI-powered apps in Replit and deploy them directly into Microsoft Fabric with enterprise-grade...

Salesforce launches AI-powered Slackbot to enhance workplace productivity

Salesforce has introduced a rebuilt Slackbot that evolves from a notification tool into a fully powered AI agent capable of searching enterprise data, drafting documents, and taking actions. This move positions Salesforce to compete more directly with Microsoft and Google in the workplace AI space.

Why it matters: The new Slackbot represents a significant step in integrating AI into workplace tools, potentially improving efficiency and collaboration. It highlights the intensifying competition among major tech companies to lead in enterprise AI solutions.

Anthropic launches Cowork, a no-code Claude Desktop AI agent for file work

Anthropic introduced Cowork, an AI agent that extends Claude Code's capabilities to non-technical users, enabling file-based work without coding. The feature was developed in about a week and a half.

Why it matters: Cowork lowers the barrier for using advanced AI tools by allowing users to interact with their files through an AI agent without needing programming skills. This could broaden AI adoption among everyday users.

LLMs and GenAI Pose Growing Risks to Cybersecurity and Platform Integrity

Large Language Models (LLMs) and generative AI systems like ChatGPT and Gemini are transforming digital platforms but also raising significant cybersecurity, privacy, and platform integrity challenges. Notably, LLM-assisted malware is projected to increase from 2% in 2021, highlighting escalating r...

Why it matters: As LLMs become more integrated into various sectors, their misuse in malware and scams threatens digital trust and safety. Understanding these risks is crucial for developing effective safeguards in AI deployment.