All news with #ai runtime security tag

🚀 Google Cloud and Anyscale describe tighter integration between Ray and Kubernetes to improve distributed AI scheduling and autoscaling on GKE. The release introduces a Ray Label Selector API (Ray v2.49) to align task, actor and placement-group placement with Kubernetes labels and GKE custom compute classes, enabling targeted placement and fallback strategies for GPUs and markets. It also adds Dynamic Resource Allocation for A4X/GB200 racks, writable cgroups for Ray resource isolation on GKE v1.34+, TPU/JAX training support via a JAXTrainer in Ray v2.49, and in-place pod resizing (Kubernetes v1.33) for vertical autoscaling and higher efficiency.

🛡️ This recap of the Agent Factory episode explains practical strategies for securing production AI agents, demonstrating attacks like prompt injection, invisible Unicode exploits, and vector DB context poisoning. It highlights Model Armor for pre- and post-inference filtering, sandboxed execution, network isolation, observability, and tool safeguards via the Agent Development Kit (ADK). The team demonstrates a secured DevOps assistant that blocks data-exfiltration attempts while preserving intended functionality and provides operational guidance on multi-agent authentication, least-privilege IAM, and compliance-ready logging.

⚙️ Microsoft is introducing Copilot Actions, a Windows 11 Copilot feature that allows AI agents to operate on local files and applications by clicking, typing, scrolling and using vision and advanced reasoning to complete multi-step tasks. The capability will roll out to Windows Insiders in Copilot Labs, extending earlier web-based actions introduced in May. Agents run in isolated Agent Workspaces tied to standard Windows accounts, are cryptographically signed, and the feature is off by default.

🔒 Google has begun rolling out an AI-powered ransomware detection feature for Google Drive for desktop. The feature automatically pauses syncing of affected files on Windows and macOS when it detects signs of ransomware, protecting cloud copies though it does not prevent local file encryption. Administrators may disable detection or file restoration via the Admin console, and alerts require Drive version 114 or later.

🔍 Join The Hacker News for a free webinar, "Workflow Clarity: Where AI Fits in Modern Automation," featuring Thomas Kinsella, Co‑founder & Chief Customer Officer at Tines. The piece argues that human-only processes are slow, rigid rule engines break when reality changes, and fully autonomous AI can create opaque, unauditable paths. Attendees will learn practical mapping of tasks to people, rules, or AI, how to spot AI overreach, and patterns for building secure, auditable workflows that scale without sacrificing control.

🚀 Google Cloud announced BigQuery performance and usability enhancements to accelerate generative AI inference. Improvements include >100x throughput for first-party text generation and >30x for embeddings, plus support for Vertex AI Provisioned Throughput and dynamic token batching to pack many rows per request. New reliability features—partial-failure mode, adaptive traffic control, and robust retries—prevent individual row failures from failing whole queries and simplify large-scale LLM workflows.

🔒 CrowdStrike explains how widespread AI adoption expands the enterprise attack surface, exposing models, data pipelines, APIs, and autonomous agents to new adversary techniques. The post argues that legacy controls and fragmented tooling are insufficient and advocates for real-time, full‑stack protections. The Falcon platform is presented as a unified solution offering telemetry, lifecycle protection, GenAI-aware data loss prevention, and agent governance to detect, prevent, and remediate AI-related threats.

🔐 CrowdStrike explains how inadvertent 'leakage' — when dependent or correlated observations are included in training — can inflate machine learning performance and undermine threat detection. The article shows that blocked or grouped data splits and blocked cross-validation produce more realistic performance estimates than random splits. It also highlights trade-offs, such as reduced predictor-space coverage and potential underfitting, and recommends careful partitioning and continuous evaluation to improve cybersecurity ML outcomes.