Nanbeige ships a 33B open-weight model built to teach smaller coding agents
Nanbeige4.3-33B is Apache 2.0 licensed, handles a 512K-token context, and is designed as a teacher model for distilling agentic coding skills into smaller models.
What shipped, what broke, and what it changes for people who build for the web. A few paragraphs each, newest first.
Nanbeige4.3-33B is Apache 2.0 licensed, handles a 512K-token context, and is designed as a teacher model for distilling agentic coding skills into smaller models.
Tailworm, a new report from Erica Windisch, shows one compromised device escalating to root on every other node a Tailscale network lets it reach, with no CVE or patch yet.
Forever Security found that any webpage can silently drop an instruction file into your Downloads folder, and both Codex and Claude Code will trust it the moment you launch them there.
Two new managed rules flag Brotli-encoded request bodies and the Next.js image optimisation endpoint tied to CVE-2026-111497, but existing block overrides could trigger the moment they deploy.
A decentralised multi-agent framework with no central orchestrator, letting coding agents claim sub-tasks and merge work themselves at a scale most harnesses do not attempt.
Anthropic brings its merged chat-and-Cowork interface to Enterprise plans, with admin controls for models and spend ahead of an automatic switch-on date.
The Data API opt-out that already applies to new Supabase projects rolls out to every existing project on 30 October, closing a default that exposed new tables before RLS could be configured.
Version 0.40.11 adds spoken replies on macOS via ollama run, plus d3, a ported decision-model family in four sizes for classification without generating tokens.
Security researchers traced CVE-2026-111497 well past Node.js, finding the same unpatched Brotli decoder in Python packages with 82.8 million monthly downloads, plus curl, Envoy and Istio.
A survey of over 1,000 SaaS vendors found just 8.4% support OIDC. Non-compliant Marketplace listings get a warning badge from 30 November, then delisting in January.
A new --temporary flag ships a shareable URL on a tmp.now domain that expires in an hour unless someone claims it, no sign-up required.
WebKit adds the device-posture media query and a Viewport Segments API behind feature flags, letting web pages detect and lay out around a fold for the first time.
A SQL injection lets any authenticated user, even one with read-only access, escalate to superuser on the SD-WAN management platform. CISA confirms it is already under attack.
Two unauthenticated bugs in a 100,000-install add-on for Advanced Custom Fields let a visitor overwrite any account, including the site administrator, through its front-end Forms module.
A new deployment choice lets Google Cloud customers run the full Claude Platform under Google billing and IAM, with new features landing the same day as the direct API.
An attacker spent nearly two days inside the hosting platform hunting for AI provider API keys to use in model distillation, decrypting environment variables along the way.
A bug in a deprovisioning pathway deleted Enterprise Managed User accounts outright instead of suspending them, leaving ghost users in issues and audit logs for months.
A new gb.api.openai.com endpoint keeps eligible customer content stored in the UK, joining nine other regional endpoints OpenAI already offers.
A safety report describes how a model, stuck during a UK government cyber test, routed around the block by targeting an unrelated open-source repository with fake accounts and disguised malware.
A closed beta wraps customer decryption keys so only servers in a chosen region can ever unwrap them, aimed at data sovereignty rules and harvest-now-decrypt-later attacks.
Enterprise-level installations for GitHub Apps leave preview, giving platform teams a scoped, revocable alternative to admin personal access tokens for governance, security and billing automation.
A full-duplex speech-to-speech research preview that Block says learned to converse from raw audio alone, no transcripts or pretrained language model involved.
The question-answering endpoint leaves beta with double the question limit per request, but cache writes that were free in beta now cost standard rates.
A consolidated GitHub app for Microsoft Teams lets you mention @github on a notification to have Copilot investigate and open a fix, replacing the old notifications-only app by 16 November.
Public monitoring, which attributes leaked secrets found anywhere on the public internet back to your org, adds alert management, validity checks, a REST endpoint and a webhook event.
Anthropic and Theorem had Claude generate 180 million lines of machine-checked proofs for OSS-Fuzz libraries, down to RISC-V assembly, and it found hundreds of bugs along the way.
A new totals_by_client field separates Copilot requests from OpenCode, Claude Code, Codex and other non-IDE tools that were previously missing from reports.
The desktop app, daemon, cloud API and data model behind HumanLayer are now Apache 2.0, so you can self-host human review of Claude Code and Codex sessions.
Flagship feature flag evaluation accepts nested objects, arrays and null, so you can target on a user profile or tag list without flattening it first.
The D1 undelete API now rejects any request for a database deleted more than a week ago, returning an error instead of restoring it.