The White House has completed its voluntary artificial intelligence oversight framework, but has not released the document or said when companies will begin using it. OpenAI, Google, Anthropic, and Meta are being invited to review the process as officials work out how frontier AI models should be tested for cybersecurity and national-security risks.

Source

Mistral is turning Europe’s push for AI sovereignty and growing distrust of closed models into a commercial opportunity. The French lab is reportedly preparing a funding round that could raise its valuation from $13.5 billion to $23 billion, while its revenue has grown 20-fold in a year.

Source

The GitHub Copilot Harness is now a stable execution engine for Microsoft Agent Framework agents in both.NET and Python. The integration combines Copilot’s coding tools with Agent Framework’s approval controls, custom tools, session handling, and OpenTelemetry observability.

Source

Microsoft is introducing AI token budget targets and spending tracking for engineers, while warning that “tokenmaxxing” is not its productivity goal. The company will make the cheaper OpenAI GPT-5.6 the default model for internal use as GitHub Copilot adoption expands.

Source

Coding agents can produce complete features quickly, but their default output is often a single, oversized pull request. GitHub is promoting stacked pull requests as a way to split that work into smaller, dependent changes that humans can review, test, and merge without manually maintaining every branch.

Source

Texas has temporarily halted new data-center grid connections while the Public Utility Commission of Texas and ERCOT audit projects seeking approval. The review comes as ERCOT’s interconnection queue reaches 474 gigawatts—more than five times the grid’s current peak demand—with roughly 90% tied to data centers.

Source

AWS Kiro Crew is an open-source orchestrator that keeps coding agents working across sessions, schedules, and repositories—even when developers are offline. Unlike a cloud-only service, Crew can run on laptops, containers, or virtual machines without an AWS account, giving teams more control over where agent activity and project data reside.

Source

Microsoft is turning Zero Trust for AI into an operational checklist with expanded assessment tools, a 91-task DevSecOps pillar, and new guidance for AI memory. The update gives security and platform teams a way to baseline AI-agent risks and build a phased remediation roadmap.

Source

ServiceNow is expanding Autonomous Security with AI Specialists that can investigate and remediate security issues with limited human intervention. The new framework unifies six security areas, including vulnerability management, identity governance, incident response, and compliance, within the ServiceNow AI Control Tower.

Source

A fast-moving Keyv-linked npm worm has spread far beyond its original namespaces, poisoning hundreds of package names while stealing credentials from developer and CI environments. Its repository payload also uses Claude Code and VS Code project hooks to trigger when developers open or trust affected workspaces.

Source

Cybercriminals have shifted from ClickFix lures to cloned GitHub repositories that impersonate popular AI tools and developer resources. The campaign uses SmartLoader to hide an infostealer capable of stealing API keys, cloud credentials, browser secrets, and CI/CD access tokens.

Source

Red Hat has launched asago, an open-source project designed to convert corporate and regulatory AI policies into production controls. The workflow connects risk mapping, testing, mitigation, deployment, and auditing, with the goal of reducing governance work from months to days.

Source

Cloudflare’s new CI SDK lets platforms run build, test, and deployment pipelines as durable Workflows instead of managing separate queues, consumers, and state. The TypeScript-based pipelines can trigger directly from Artifacts repository pushes, run checks in isolated environments, and use an AI agent to create fixes when a build fails.

Source

Cloudflare is launching Cloudflare Agents with agent-specific observability rather than treating AI workloads like ordinary applications. The new dashboard traces model calls, tool use, subagents, approvals, token consumption, and underlying Workers activity, while session replay shows the context behind each decision.

Source

Cloudflare Wallets is moving AI agents closer to autonomous online purchases. The new service lets account owners create agent-controlled Virtual Wallets with spending caps, allow lists, and transaction limits, while Cloudflare Wallet handles provide a persistent identity for machine-driven transactions.

Source

A new Azure operations assistant demonstrates a safer pattern for agentic IT automation: read-only diagnostics run immediately, while incident creation remains pending until the authenticated requester explicitly confirms it. The reference implementation also exposes a production concern many AI demos hide—its approval state is in memory and cannot safely scale beyond one Container Apps replica.

Source

OpenAI is challenging Apple’s trade-secrets lawsuit, saying Apple’s lawyers contacted the wrong person, misrepresented the follow-up, and waited five months before filing. It also argues that Apple’s own offboarding failures left former employees with access to confidential files.

Source