Top line

Cursor adds realtime voice chat with agents

Cursor's Nov. 6 changelog introduces Voice Chat with Agents: a two-way realtime voice call so you can describe a change out loud, ask what an agent is doing, and hear progress while it works. It runs in the Agents Window, the IDE, on the web, and in Cursor for iOS. On iOS you can start a call with "Hey Siri, call Cursor" and mute or hang up from a Lock Screen Live Activity. Voice chat is separate from dictation (the older Voice Mode); it is powered by Grok Voice with selectable voices, and Cursor says Privacy Mode runs voice chat with zero data retention. Update Cursor to use it.

Why it matters: Steering agents by voice changes how teams interrupt and monitor long runs without typing, including from a phone. Treat it as a new interaction surface, not a model upgrade, and confirm Privacy Mode policy before enabling it on regulated work.

Sources: Cursor Voice Chat changelog · Cursor changelog

xAI turns on default Grok API text watermarking

SpaceXSI docs say that starting Nov. 7, 2026, text responses from grok-4.7, grok-4.6, grok-4.5, grok-4.3, and the grok-4.20 reasoning, non-reasoning, and multi-agent variants include an invisible watermark by default on every SpaceXSI API endpoint, including Responses and Chat Completions (structured outputs and tool-call arguments included). There is no request parameter; check the x-watermark response header. Team admins can opt out per model in the Console (changes apply within about five minutes), with some models possibly locked on. Docs frame the feature as EU AI Act compliance.

Why it matters: Downstream classifiers, eval harnesses, and disclosure workflows may start seeing watermarked text without code changes. Admins who need unmarked output should opt out before relying on the old behavior, and document any legal basis they assert when opting out.

Sources: Output Watermarking docs

OpenRouter rebuilds Auto Router and launches Auto Code

OpenRouter's Nov. 6 announcement ships a rebuilt Auto Router (openrouter/auto) and a coding-specialized Auto Code (openrouter/auto-code). OpenRouter says the new stack outperforms the prior router by up to 4× on its benchmarks, with a routing step about 3× faster at p90 (0.65s vs 1.96s). Routing weighs eleven request factors plus live price, provider health, and cache status, and respects guardrails, ZDR, allowlists, and BYOK. Auto Code is tuned for harnesses; OpenRouter claims scores in line with Claude Opus 5.5 at about 74% lower cost per task. Existing openrouter/auto callers need no slug change; set cost_tier=low to restore the older cheap bias.

Why it matters: Teams that pinned Auto for mixed or coding traffic get a different model mix and latency profile overnight. Re-check cost dashboards and harness quality before assuming yesterday's routing still holds.

Sources: OpenRouter Auto announcement

What to do

  • Update Cursor and try Voice Chat where policy allows; confirm Privacy Mode for voice. Cursor Voice Chat changelog
  • Review xAI Console watermarking defaults for each Grok text model your team uses. Output Watermarking docs
  • Rename gemini-embedding-2-preview to gemini-embedding-2 in clients (same schema, pricing, and embedding space; no re-index). Gemini API changelog
  • Deploy ChromeOS / ChromeOS Flex 16836.32.0 (Browser 156.0.8078.93) for the Nov. 6 High fixes. ChromeOS security update
  • Keep hunting Apple devices below the May 2024 WebKit fixes. CISA's due date for CVE-2024-27834 remains Nov. 18. Apple iOS 17.5 notes
  • OpenAI prepaid API users on some U.S. business Mastercards: add a backup card if auto-recharges fail (listed BINs on status). OpenAI status

Also notable

  • Gemini Embedding 2 Preview is deprecated; effective Nov. 6, preview requests route to gemini-embedding-2. Gemini API changelog · Gemini deprecations
  • GitHub Copilot code review can now be managed through REST and GraphQL from enterprise policy down to repos, with compact overviews and auto-retry/failover (Pro through Enterprise). Copilot code review changelog
  • @GitHub is generally available in Slack and Teams, including Canvas in Teams. GitHub Slack/Teams GA
  • Kimi K3.1 Pro (Highspeed) is on Amazon Bedrock in all Bedrock Regions via cross-Region inference, with prompt caching. AWS What's New
  • Claude Enterprise (beta): eligible orgs can tag projects with cost or client-matter codes and see most spend by tag. Claude release notes
  • Claude Code v2.1.319 tightens auto-mode when the server cannot check a call (ask/deny, no built-in classifier fallback). Claude Code 2.1.319
  • vLLM publishes LibTorch Stable ABI CUDA wheels as of v0.33.0. vLLM Stable ABI post
  • CISA KEV: SonicWall SonicOS SSLVPN CVE-2026-19868 added Nov. 6 (due Nov. 20). CISA KEV
  • Watch: Hy4 full open weights and the Pentagon AI-lab legal inventory remain unpublished; WordPress KEV CVE-2026-119977 federal due date was Nov. 6; GitHub's Nov. 5 Actions RCA is still pending.
  • Codex / ChatGPT limits: Tibo Sottiaux posted no new resets or limit changes in-window. The Oct. 29 Pro doubling and the December shared ChatGPT/Codex usage plan still stand. Tibo on X

Sources

Cursor Voice Chat changelog · Cursor changelog · Output Watermarking docs · OpenRouter Auto announcement · Gemini API changelog · Gemini deprecations · ChromeOS security update · Copilot code review changelog · GitHub Slack/Teams GA · AWS What's New · Claude release notes · Claude Code 2.1.319 · vLLM Stable ABI post · CISA KEV · OpenAI status · Apple iOS 17.5 notes · Tibo on X