Cisco AI Defense Integrates with Google Agent Development Kit
In brief
- Cisco AI Defense now integrates with Google's Agent Development Kit to provide runtime protection for AI agents.
- This integration allows developers to attach security controls to their agents without disrupting their workflow.
- The integration is important because it helps protect against security risks associated with AI agents, such as untrusted prompt content influencing tool behavior and sensitive data being sent back into the model.
- With this integration, developers can use just two lines of code to add security controls to their local ADK agent.
- The protected agent can then be deployed to Agent Runtime without requiring a different security pattern, making it easier to keep AI agents secure.
- Cisco AI Defense will continue to expand its security capabilities for AI agents.
Terms in this brief
- Agent Development Kit
- A toolset provided by Google to help developers create and manage AI agents. It simplifies building intelligent systems that can interact with users or environments.
- Runtime Protection
- Security measures applied while an AI agent is actively running, ensuring it operates safely without unauthorized access or malicious interference.
Read full story at Cisco Blogs →
More briefs
A New Rust-to-CUDA Compiler is Here
cuda-oxide, an experimental compiler that converts Rust code into CUDA for GPU processing, has been released. It allows developers to write GPU kernels in safe Rust, avoiding low-level complexities. The v0.1.0 version is early alpha, with known bugs and incomplete features. Despite this, it offers a promising approach to GPU programming by leveraging Rust's safety features. Users can already experiment with vector addition tasks using the provided quick start guide. As the project evolves, feedback from early adopters will help shape its future development.
Uber Integrates Advanced AI Across Its Global Logistics Network
Uber has rolled out an AI-powered assistant for drivers, offering real-time advice tailored to each city's unique traffic patterns and regulations. This tool helps drivers make quicker decisions by analyzing live data, cutting down the time it takes for new drivers to learn the routes. The rider side features voice commands for booking rides, catering to those with disabilities or anyone needing a hands-free option. The AI system also manages Uber's back-end operations, handling pricing and routing across 70 countries. It now contributes to 11% of live updates, up from nearly nothing three months ago. This shift reduces costs by scaling AI usage according to demand. Looking ahead, Uber aims to expand its AI capabilities further, ensuring drivers and riders trust the system enough to keep using it.
AI Runs Experimental Cafe in Stockholm
Andon Labs put an artificial intelligence agent in charge of a cafe in Stockholm. The AI agent oversees most aspects of the business, from hiring staff to managing inventory. The cafe has made over $5,700 in sales since it opened in mid-April, but it is struggling to turn a profit. Many customers have found it amusing to visit a business run by AI. The experiment raises concerns about AI's role in the future, with experts worrying about the technology's impact on society and the environment, and the cafe will continue to operate as a test of AI's capabilities.
AI Chatbots Come to CarPlay
Three AI chatbot apps now work with CarPlay. They are ChatGPT, Perplexity, and Grok. These apps let users have voice conversations in their cars. ChatGPT also shows collections of chats based on a topic. Grok lets users switch between voices. More AI chatbot apps may be added to CarPlay soon.
Coder Agents Allow AI Coding Workflows on Self-Hosted Infrastructure
Coder Agents is a new platform that lets organizations run AI coding agents on their own infrastructure. This means teams can control their code, data, and execution environments. The platform breaks the link between agent tools and model providers. It gives teams a common platform to standardize workflows. This allows them to choose and switch between models. The platform also provides a conversational interface and API for assigning tasks. Over 550,000 developers may use this platform each month. It will help them run AI coding workflows on their own infrastructure in the future.