AI news daily brief: 2026-11-13
Nine stories today: five product launches, one security incident, one leadership change, and two model releases. Three already have full standalone treatment on Pondero and one has a dedicated guide; this brief covers all nine at a glance with links to the deeper versions.
Google's Antigravity ships a Swarm mode that puts agent-to-agent chatter on screen
Google's Antigravity coding IDE shipped a /swarm command on November 9, 2026 that splits a task across named agents working in parallel and coordinating over one shared, watchable message board, per Antigravity's own documentation. A coordinator assigns workers, a verifier checks diffs and tests, and a monitor wakes only if the team stalls; idle members burn no quota until mentioned. Google warns that because every worker is its own agent, a swarm run consumes quota "much faster than a single conversation." TestingCatalog reported the live dashboard on November 10, noting that rival tools from Kimi, Meta, and xAI keep this kind of multi-agent coordination hidden from the user by default. Full story: Google ships a Swarm command for Antigravity that exposes agents' coordination messages live.
Claude Code 2.1.323 lets Claude write JavaScript to drive its own tools
Anthropic released Claude Code 2.1.323 on November 12, 2026 with an experimental, opt-in "Code Mode" where Claude writes JavaScript to call its own tools instead of issuing direct tool calls, per the official changelog. The same release adds .docx support for Artifact asset uploads, routes Claude-Tag sessions started in Slack into the Enterprise Compliance API so admins can list and read session transcripts, and adds a CSV-export button for per-repository, per-org, or catalog-wide code-review costs broken out by day, week, or month. Full story: Claude Code's Code Mode: what it is and whether to turn it on.
A Fireworks AI breach exposed Notion's AI chat transcripts
Notion disabled its AI-powered Insights feature and cut ties with inference vendor Fireworks AI after confirming a breach there exposed chat transcripts tied to that feature, per Notion's security advisory. Fireworks first flagged unusual access on November 2, disclosed actual customer exposure on November 6, and confirmed the final scope on November 8; Notion says fewer than 1% of workspaces were affected and most exposed transcripts are "unattributable" to a specific user. The root cause traces back to a breach at the cloud provider hosting Fireworks' own research machines, disclosed separately in late October. Affected workspace owners can pull a CSV of exactly what was exposed from their sidebar. Full story: A Fireworks AI breach spilled Notion's AI chat transcripts into attackers' hands.
GitHub lets a GitHub App manage Enterprise Copilot seats without a personal token
GitHub shipped a new fine-grained permission, Enterprise Copilot license management, on November 10, 2026, letting a GitHub App call the enterprise-level Copilot seat APIs to assign, remove, and list seats for users and enterprise teams, per GitHub's changelog. Admins previously had to script seat changes against a personal access token tied to one person's account, which breaks if that person leaves or rotates credentials; an app-scoped integration survives staff turnover. Full story: Manage GitHub Copilot seats with a GitHub App instead of a personal token.
Jensen Huang resigns as Nvidia CEO, Colette Kress named interim chief
Nvidia's board accepted Jensen Huang's resignation as president, CEO, and director effective immediately, and named CFO Colette Kress interim chief executive while she keeps her finance title, per a Form 8-K Nvidia filed with the SEC on November 12, 2026. The board also named Harvey C. Jones, former CEO and executive chairman of Synopsys, as the company's first-ever board chair. Huang had been Nvidia's most visible advocate for open-weight models, tied to its Nemotron family and its pending Hugging Face acquisition; neither Kress nor Jones addressed open-source strategy specifically in the filing, per TechCrunch. Nvidia's Q3 fiscal 2027 earnings call goes ahead as scheduled on November 17. Full story: Jensen Huang resigns as Nvidia CEO, Colette Kress named interim chief.
Google turns on Gemini email summaries and drafting for every Workspace tier
Google flipped AI Overview conversation summaries and Help me write to on by default across every paid Workspace edition on November 10, 2026, not just the top-tier plans that carried them before, per Google's Workspace Updates post. The rollout reaches Workspace Individual subscribers and personal accounts for users 18 and older too, taking up to 15 days across Rapid Release and Scheduled Release domains. Because both features ship on by default, organizations that previously avoided Gemini by staying on a lower-tier edition inherit it automatically and now have to actively switch it off rather than opt in. Full story: Google turns on Gemini email summaries and drafting for all Workspace tiers.
Meta's Muse Ember matches Claude Haiku 5.5's score at a third of the cost per task
Meta released Muse Ember on November 9, 2026, its smallest model yet, built for high-volume agent subtasks like tool calls, coding, routing, and parsing rather than frontier reasoning. It prices at $0.10 per million input tokens and $0.50 per million output tokens, and scores 36 on Artificial Analysis's Intelligence Index, matching Claude Haiku 5.5's score while costing about $0.14 per task against Haiku's roughly $0.41, per Artificial Analysis. The gap comes from Ember writing fewer tokens to reach the same answer, not from a cheaper rate card; Haiku 5.5 lists the same per-token price. Ember is available now through Meta's first-party API and Muse Code, paired with the larger Muse Wonder model for heavier steps in the same pipeline.
DeepMind ships Gemini Robotics 2.5 with an open jailbreak-safety benchmark for robots
Google DeepMind introduced Gemini Robotics 2.5 on November 10, 2026, calling it its most capable vision-language-action model yet for robot dexterity and reliability, and began rolling it out to hardware partners including Apptronik, Agile, and Mercedes-Benz, per DeepMind's blog post. Alongside the model, DeepMind released ASIMOV-Jailbreak, an open benchmark on Hugging Face that tests robots against hundreds of deceptive prompts disguising unsafe requests across dozens of scenes, letting outside teams check embodied-agent safety without waiting on DeepMind's own evaluations.
Workspace Studio adds a Wait until step so automated flows can pause for up to 14 days
Google Workspace Studio began rolling out a Wait until step on November 9, 2026 that pauses a flow anywhere in its execution, including inside conditional branches and loops, and resumes it at a set day and time or after a duration of up to 14 days, per Google's Workspace Updates post. Before this, delaying a sequence meant splitting it into separate scheduled flows; the new step keeps multi-stage automations, like a delayed follow-up email or a Monday-morning Chat digest, inside one flow. It reaches Business, Enterprise, and Education tiers with select add-ons over one to three days on both release tracks.
Sources
- Swarm (/swarm): Antigravity Docs: Google, feature documentation
- Claude Code changelog: Anthropic, November 12, 2026
- Fireworks AI incident: Notion Security Bulletin: Notion security advisory
- GitHub App support for Enterprise Copilot seat management: GitHub changelog, November 10, 2026
- NVIDIA Announces CEO Transition (Form 8-K, Exhibit 99.1): SEC filing, November 12, 2026
- Expanding Gemini in Gmail to more Google Workspace users: Google Workspace Updates blog, November 10, 2026
- Meta's Muse Ember is the cheapest model scoring 36 on the Intelligence Index: Artificial Analysis, November 9, 2026
- Gemini Robotics 2.5 accelerates real-world physical intelligence: Google DeepMind blog, November 10, 2026
- Pause and schedule automated workflows with the new Wait until step in Workspace Studio: Google Workspace Updates blog, November 10, 2026
