Apple opens Siri AI integration to every app developer
A new Developer Preview lets any app expose its content and actions to Siri through App Intents, with App Store submissions opening in spring.
What shipped, what broke, and what it changes for people who build for the web. A few paragraphs each, newest first.
A new Developer Preview lets any app expose its content and actions to Siri through App Intents, with App Store submissions opening in spring.
A new public beta reviews each GitHub push for vulnerabilities and can hand fixes to Vercel Agent. Free until 10 November, then billed per token.
GitHub extends automatic dependency updates to apm.yml, the manifest format for AI agent skills, instructions and MCP servers.
Queue::whenJobRunsLongerThan and Schedule::whenTaskRunsLongerThan give Laravel apps a hook for catching jobs and scheduled tasks creeping toward a timeout, something the framework never offered before.
Next.js 16.4.5 and 15.5.29 fix seven moderate-severity issues, the sharpest of which lets client-side JavaScript read HttpOnly session cookies on Edge runtime routes.
Copilot joins JetBrains Air's agent picker, running tasks locally through your IDE or in an isolated cloud container via Air Teams, billed against your existing Copilot subscription.
A new Inertia::live() helper refreshes a page prop whenever a broadcast event fires, with transports for Laravel Echo, Pusher and Action Cable.
A 38,675-page benchmark and a cheap 0.8B-parameter detector for catching prompt injection in browser agents, though Perplexity's own adaptive-attack tests show a determined attacker still gets through most of the time.
FlashProve is an open-source zkSNARK prototype that cryptographically proves a reply came from specific committed model weights, at roughly 70 to 95 times the GPU cost of running the model itself.
Agents SDK 0.33.0 batches storage writes instead of writing on every update, cutting SQLite transactions up to 27x, but a conversation written on Pi 1.5 cannot be read back by Pi 1.4.
Zero Wire on Cloudflare Radar turns critical vulnerabilities into short briefings with remediation steps, free for anyone, not just Cloudflare customers.
A resumed tool call could reach your approval policy with a stripped-down tool definition, missing the metadata the same call carried the first time round.
The /crawl endpoint now returns a per-URL outcome and a robots.txt diagnostics array, replacing a bare success-or-failure result with an actual reason.
Eligible accounts on GitHub Desktop 3.6.10 and later now get automatically refreshing tokens instead of long-lived ones, shrinking how long a leaked credential stays useful.
A major CLI release changes which project your commands hit by default, requires confirmation before touching a hosted project, and drops the supabase-go binary.
A flaw fixed back in August is now being exploited against self-hosted BTCPay Server instances still on 2.4.1 or earlier, draining Bitcoin and Litecoin from hot wallets since 1 November.
Codex-driven commits grew far faster than stock Git could serve, so OpenAI built Pando, a Git-compatible backend now handling nearly a petabyte of data a day.
Service tokens can now carry up to five key-value tags, and Service Auth policies can authorize by tag instead of listing every token by name.
Usage charges for the RAG service were due to start on 1 November. Cloudflare has pushed that to 1 December and doubled the default sync interval to 24 hours.
The renamed "AI Crawler and Agent access policies" setting lets you allow or block the top Personal Agents one by one, instead of one blanket rule for all of them.
Version 2.1.317, shipped 3 November, asks for confirmation before a web or shell tool call sends your account email address to the network, and adds a cost report API for gateway admins.
A TOML alternative to vercel.json ships for every project, with comments and a flatter structure. You pick one config file, not several.
A new beta parameter on the Messages API keeps a conversation server-side, so each turn in a long agent loop sends only what changed.
From 2 November both models are gone from Google AI Studio. From 1 December, API requests to them route automatically to newer models, with breaking parameter changes.
Vercel MCP now accepts Client ID Metadata Documents, so any MCP client can connect without Vercel reviewing or allowlisting it first. The consent screen shows the client is unverified.
Builds on every plan automatically fail and try to invalidate their cache if a dependency matches a blocklist of malware or vulnerable package versions, no setup required.
From 3 November, Opus 4.7, Opus 4.6, Sonnet 4.6 and Mythos 5 carry the same imperceptible mark Anthropic has used on new models since August.
A mixture-of-experts model with 48 billion active parameters, a 1-million-token context window and native image handling, aimed at long-running coding and design agents rather than chat.
The property finally takes a custom string instead of the default ellipsis when text is clipped, something Firefox has supported since 2011.
The first MCP App from Vercel's own MCP server renders live build status, logs and a Visit button directly inside ChatGPT and other MCP App clients.