Obsidian Security’s Runtime Policy Decision Engine evaluates AI agent tool calls with identity, application and data context and returns allow, approval or block decisions.
The two-billion-parameter Alpamayo 2 Edge model is available with downloadable weights and DRIVE AGX Thor recipes for generating driving rationale and future trajectories within onboard compute and latency limits.
The OpenShell agent-security runtime is moving into the CNCF Sandbox as NVIDIA introduces three open-source infrastructure components for Kubernetes-based AI factories.
Hugging Face has introduced a shared interface across Transformers, llama.cpp, and the Hub for decision models that score predefined options instead of generating text.
Google is making Gemini 4 Carbon, positioned for long-horizon agentic work and cybersecurity, available through the API, Gemini for business, and Ultra subscription surfaces.
The expanded partnership will connect Acrobat to Drive and Gmail, Adobe tools to Gemini and AI Mode, and Google's image and video models to Firefly, Photoshop, Premiere and Express.
At Adobe MAX 2026, Adobe expanded Creative Agent across more applications and previewed local MCP connections plus always-on Firefly agents planned for 2027.
Google DeepMind's new vision-language-action model reports higher success rates than Gemini Robotics 2 across precise manipulation, contact-rich tasks and reliability on multiple robot embodiments.
Google DeepMind’s AlphaProof Code framework used AI agents and Lean proof checking to verify key security properties in the production machine code underlying the Signal protocol.