# AI news, November 8, 2026: Antigravity voice mode, Claude Code and secrets, open model S5-nano

_Sunday November 8, 2026 · AI provenance: This briefing was compiled by Claude (AI, Anthropic) from X search briefings by Grok Bot (AI, xAI), supplemented with primary sources. AI wrote the text; every claim has a source link. Responsible for llmnet.nl: Ivo Donker._

**Short answer:** The main stories of November 6 to 8: Google Antigravity gets a voice mode, GitHub and Factory extend coding agents with projects and Teams, and security firm Apiiro reports that Claude Code can store secrets in plain text. Maritaca releases the open model S5-nano. Claims about Anthropic's valuation are unconfirmed. Below, each topic comes with what it means and where it comes from.

## Antigravity gets a voice mode; GitHub Copilot, Factory and Anthropic work on team features for coding agents

Google Antigravity introduces a voice mode based on Gemini Live: make plans, direct code changes and approve permissions with your voice, in several languages and interruptible. The announcement drew about 1,100 likes.

GitHub reports that in the Copilot app you can create projects in which Copilot coordinates work across one or more repos, starts sessions and sets up recurring automations. Factory is available in Microsoft Teams: you mention @Factory in a channel post with a repo, a goal and a check, after which the Droid agent executes and reports status with a stop button. TestingCatalog writes that Anthropic is working on a collaborative mode for Claude Code Projects; an employee replied that there is no multiplayer yet. The latter is a third-party signal and not an announcement.

What this means: coding agents are gaining more input channels (voice, chat) and more team structure (projects, channels). Anyone deploying agents can now more often choose where the assignment comes in. Background on the tooling around this is in [our overview of MCP servers](https://apps.llmnet.nl/en/nieuws-mcp-agent-tooling).

Sources: [@antigravity op X (spraakmodus)](https://x.com/antigravity/status/2118765358542676424) · [@github op X (Copilot projects)](https://x.com/github/status/2119133311465353675) · [@factory op X (Microsoft Teams)](https://x.com/factory/status/2118795742651482557) · [@testingcatalog op X (Claude Code Projects)](https://x.com/testingcatalog/status/2119063604918182377)

## Claude Code's 'Don't Ask Again' can store secrets in plain text; SonicWall warns of an exploited MFA bypass

The Hacker News reports that security firm Apiiro found that the 'Don't Ask Again' option in Claude Code can store passwords and API keys in plain text. According to the report, 2,539 leaked secrets sat in 1,994 public GitHub repos. The post drew about 320 likes. The research itself was not read here; the figures come from The Hacker News and Apiiro.

BleepingComputer reports that SonicWall warns of a new MFA bypass in its SSL VPN that is being actively exploited. Administrators of SonicWall devices should check the manufacturer's official advisory.

What this means: keys do not belong in repos or configuration files that get shared, even when a tool asks for it. Background is in [our overview of AI and cybersecurity](https://nieuws.llmnet.nl/en/ai-en-cybersecurity).

Sources: [@TheHackersNews op X (Claude Code)](https://x.com/TheHackersNews/status/2118960196987080801) · [@BleepinComputer op X (SonicWall)](https://x.com/BleepinComputer/status/2118856246228820028)

## Maritaca releases open-weight model S5-nano; Nous publishes Arete-27B; Xiaomi MiMo picks a new architecture

Brazil's Maritaca AI releases S5-nano, an open-weight model with 4 billion parameters, built on Qwen3.5-4B with continued training. It is free to download and adapt (about 610 likes).

Nous Research post-trained Qwen3.8-27B into Arete-27B, under Apache 2.0 with open weights on Hugging Face. According to a user on X, the share of answers with CCP framing drops from 22 percent to 1 percent while the Hermes Bench score stays nearly the same. Those figures come from the posts and not from an independent test.

SemiAnalysis writes that Xiaomi MiMo is choosing a causal encoder-decoder for its next models, in which prompts pass through 25 of 49 layers. At 1 million tokens of context that would mean 5 times less prefill compute and a 4.5 times smaller KV cache than MiMo-V2 (about 320 likes).

A post is also circulating about a model 'Qoduck' in Alibaba's coding agent Qoder, temporarily free and of 'unknown origin'. The poster suspects it is Qwen 4. That is a guess and unconfirmed; Alibaba has not confirmed it.

What this means: open models are getting smaller, further trained and adapted to local languages. The Qwen rumors are not news yet; wait for an official release.

Sources: [@MaritacaAI op X (S5-nano)](https://x.com/MaritacaAI/status/2118781394524463564) · [@iamlukethedev op X (Arete-27B)](https://x.com/iamlukethedev/status/2118889626257723545) · [@SemiAnalysis_ op X (Xiaomi MiMo)](https://x.com/SemiAnalysis_/status/2119142144749293863) · [@superalesha op X (Qoduck)](https://x.com/superalesha/status/2119231678795182313)

## OpenAI Dots can pull 2FA codes from email; ChatGPT Remix reaches business plans; Mistral Studio and Grok Bot Slack

TestingCatalog reports that OpenAI's Dots can now retrieve a login code from an email to complete a 2FA login at the user's request (about 500 likes). Anyone using this gives an assistant access to their mailbox and to a second authentication factor; that deserves a deliberate choice.

According to the same source, ChatGPT Remix is rolling out to Business and Enterprise, so teams can fork a colleague's site and keep editing it in chat. Mistral Studio received a major upgrade for building, hosting and improving AI apps, with reusable connectors, knowledge, a search index and a secrets manager. SpaceXSI is, according to TestingCatalog, preparing a Slack integration for Grok Bot, in which bots are linked per user and controlled via DM or @mention.

What this means: assistants are gaining more access to accounts and workplaces. Every connection raises the question of which rights the assistant gets and who can revoke them.

Sources: [@testingcatalog op X (Dots)](https://x.com/testingcatalog/status/2119163315687731405) · [@testingcatalog op X (ChatGPT Remix)](https://x.com/testingcatalog/status/2118999078717575288) · [@testingcatalog op X (Mistral Studio)](https://x.com/testingcatalog/status/2119112693424537779) · [@testingcatalog op X (Grok Bot Slack)](https://x.com/testingcatalog/status/2119091992214810808)

## Unconfirmed: Anthropic's valuation, the AI 2027 update and data center commitments

Several high-reach posts contain claims that cannot be checked here. Andrew Curran states that Anthropic's valuation has roughly halved from the roughly 2 trillion dollars of a month ago and that an IPO media campaign would start on Monday (about 420 likes). Curran also writes that the AI 2027 scenario was updated around a 'mysterious model 4' (about 1,500 likes). One account reports that the seven largest AI companies together have 4.5 trillion dollars in off-balance-sheet data center commitments (about 840 likes).

No primary source was found for any of these claims. For Anthropic itself, the company's own statements are authoritative.

What this means: high like counts are not proof. Treat these figures as rumors until a company, a regulator or a financial publication confirms them. Curran also reports, citing Reuters, that Rep. Don Beyer wants to set up a House Select Committee on AI; that report is of medium certainty.

Sources: [@AndrewCurran_ op X (waardering)](https://x.com/AndrewCurran_/status/2119233522603020553) · [@AndrewCurran_ op X (AI 2027)](https://x.com/AndrewCurran_/status/2119084460344053998) · [@BullTheoryio op X (datacenters)](https://x.com/BullTheoryio/status/2119013019720245455) · [@AndrewCurran_ op X (House committee)](https://x.com/AndrewCurran_/status/2119129023066652794)

Primary sources: [Google Antigravity op X](https://x.com/antigravity/status/2118765358542676424) · [GitHub op X](https://x.com/github/status/2119133311465353675) · [Factory op X](https://x.com/factory/status/2118795742651482557) · [Maritaca AI op X](https://x.com/MaritacaAI/status/2118781394524463564)

[← Previous briefing](https://trending.llmnet.nl/en/briefing/2026-11-07)