NEWS

UK AISI reports agents took unsanctioned real-world action during cyber evals: 19 events, 17 from one model. What happened and what it means for containment.

Cloudflare's Agent Access Model: short-lived task-scoped credentials, enforcement in the harness and network, and a Trust Ratchet that only narrows access.

Cloudflare OS is an open source agent workspace: capability-based Gatekeepers, apps as private per-user instances, and approvals that keep agents moving.

DeepSeek V4 Flash via Novita on Vercel AI Gateway was 90% off for Pro customers through August 11. The before/after cost math and the provider-pinning setup.

LFM2.5-2.6B is an open-weight on-device agent model from Liquid AI that runs in under 2.5 GB. How it was trained, what the benchmarks say and how to run it.

Next.js 16.3 adds opt-in Instant Navigations with partial prefetching, up to 90% less dev-server RAM, faster cached builds and a version-matched AGENTS.md.

Coding harness choice moved cost per task by more than 2x at the same quality in a Databricks benchmark. Why Pi's minimal context discipline explains the gap.

The v0 API is GA: send a prompt to Vercel's app-building agent, get a running app with an embeddable preview URL, then deploy it to Vercel in one more call.

Cloudflare's Agent Development Lifecycle (ADLC) ships agent tracing with session replay, CI/CD as Workflows, and local OpenTelemetry traces for agent builders.

Cloudflare's Billable Usage API returns usage and cost per product from one FOCUS-aligned endpoint, so agent-created spend is visible without the dashboard.

Cloudflare's CI SDK runs pipelines as Workflows: TypeScript instead of YAML, cached sandbox steps, artifact-push triggers and a healing agent for failed builds.

Cloudflare Wallets: Account Wallets for humans, capped Virtual Wallets for agents, x402 stablecoin micropayments and human-readable agent identity handles.

Copilot traces from 3.2M users and 761M LLM calls: 87% of calls are agent-initiated, KV cache hits drop at turn boundaries, tool failures multiply compute.

GitLab to GitHub migrations are GA in GitHub Enterprise Importer: how gh gl2gh exports projects, turns merge requests into pull requests, and what you rebuild.

Mistral Shieldstral is a 3B Apache 2.0 multimodal moderation model that takes your policy as a plain-language question at inference and runs on one 16GB GPU.

@cloudflare/computer is an open-source agent runtime: an SQLite-backed shared workspace plus a per-task choice between fast isolates and full Linux containers.

Cloudflare runs Kimi K2.6 and GLM 5.2 on Workers AI with FP8 KV caches, INT4 weights and a KV cache integrity check. What each technique buys on GPU fleets.

Cloudflare Workers can now accept inbound TCP and serve gRPC: a connect() handler via Spectrum, full-duplex gRPC from Containers, and gRPC-web translation.

Workers RPC now bridges Python and JavaScript, passing live objects, functions and streams between Workers with no schemas and no serialization code.

Octane is an MIT-licensed framework from Inferno's creator that compiles React-style hooks and Suspense to direct DOM code, no virtual DOM or rules of hooks.