Google: a single Gemini agent for work that can now also use Anthropic’s Claude models. At Google Cloud’s Gemini at Work event on Thursday, Google unveiled a “universal” Gemini agent built into the Gemini Enterprise app. You can give it not just instructions but goals: it plans the work itself, hands off subtasks to sub-agents and, because it runs in the cloud, is available with the same context on your phone, desktop, command line, Google Workspace, Microsoft 365, ServiceNow and Slack. Like a colleague, it gets its own Workspace account and email address, and keeps a log of its actions under its own name. It can connect to business systems (such as Jira, Confluence, BigQuery, Snowflake and Postgres) and to any MCP server. By default it picks the best model for the task itself, but users can switch — starting with Anthropic’s Claude models. For now it is available to enterprise customers in a private preview, with a consumer version coming later. According to Sundar Pichai, Gemini has more than 1 billion monthly users, and nearly 90% of Fortune 100 companies also use Gemini Enterprise. On the same day, Google released Google AI Edge Foresight, a fully offline note-taking app for Mac that records meetings using EmbeddingGemma 2 and Gemma 4 models running on the device. Source: Google Cloud, TechCrunch, The Verge, TechCrunch (Foresight).
Claude Dashboards and Claude Motion: live dashboards and animated explainers in Claude. Anthropic released both new features in beta on Thursday. Connected to data platforms (Redshift, BigQuery, ClickHouse, Databricks, Snowflake) and tools such as Salesforce, Dashboards builds continuously updating dashboards from questions asked in plain language; every chart shows when it was last refreshed, and you can inspect the query behind it. Motion turns reports, charts and product demos into short animated explainers — written as code rather than made with a video generation model, so there is no generated footage or synthetic people — which can be downloaded as MP4. Dashboards is available on all paid plans, Motion on Team and Enterprise plans; Docs, Slides and Design are out of beta and now also available on the free plan. Source: Claude, Claude (X), TestingCatalog.
Anthropic: Cyber Mission for critical infrastructure and open source, $150 million for the Genesis Mission. The new long-term program launches with two areas. The Critical Infrastructure Defense Program provides frontier Claude models, embedded engineers and threat research to providers that protect the industrial control systems of power grids, water utilities and transportation networks — these often cannot be taken offline to patch, so known vulnerabilities can stay open for years. Founding partners include Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Palo Alto Networks and Rockwell Automation. OSS Scanner, meanwhile, gives open-source projects that apply free, regular security scans with the company’s most capable models (including Claude Mythos); the reports are produced by the model without human review, so they may also contain errors. According to Anthropic, more than 29,000 potential vulnerabilities were found over the past six months, but only around 6,000 could be reviewed manually. On the same day, the company committed $150 million over three years to the US government’s Genesis Mission science program: it will bring Claude to more than 15 federal agencies (including NASA, NIH and NSF) and provide Claude, Claude Code and API credits to hundreds of research projects. Source: Anthropic (Cyber Mission), Anthropic (OSS Scanner), Anthropic (Genesis Mission).
Anthropic: a new usage policy — abusing Claude and interfering with elections are banned. The company published this year’s update to its policy, which takes effect on November 12. The new provision that has drawn the most attention bans sustained, gratuitous, abusive or cruel behavior toward Claude. Anthropic says this applies only to extreme cases, not to ordinary user frustration, arguments, dark creative themes or model testing, and that the main tool is ending the conversation (since August, Claude has already been able to leave persistently abusive conversations). A dedicated section of the new policy bans deceiving voters and disrupting elections, prohibits deceptive campaigns run with fake accounts or fabricated news sites, introduces new rules on surveillance and weapons development, and sets out specific controls for cases where Claude carries out physical actions on its own. Source: Anthropic, The Verge, TechCrunch.
OpenAI: Ultrafast mode for GPT-6.1 Sol — up to eight times faster. On Thursday, OpenAI made its new, fastest serving tier available for the GPT-6.1 Sol model in the API, Codex and ChatGPT Work. The company describes it as “near-Astra intelligence” at up to 8x the speed of standard Sol. In the API it is available to all developers, with separate usage limits, global processing, and US and EU data residency. According to TestingCatalog, it costs $12 per 1 million input tokens and $60 per 1 million output tokens. Source: OpenAI Developers (X), OpenAI changelog, TestingCatalog (X).
OpenAI: revenue is “approaching $50 billion” — not $70 billion, as previously reported — and AI stocks fell. According to the Financial Times, OpenAI told its investors that its annualized revenue was approaching $50 billion at the end of September; that is about $20 billion less than the figure widely cited at the end of September. According to a CNBC source, the earlier figure also included partners’ gross revenue, to make it comparable with Anthropic’s calculation (Anthropic counts sales made through its cloud partners, while OpenAI does not). According to the investor materials, annualized revenue grew 77% in the third quarter, and 107% in the enterprise business. Several AI stocks fell on the news on Thursday: Nvidia lost 3%, Oracle nearly 6% and CoreWeave nearly 8%. Source: Financial Times, CNBC, TechCrunch.
OpenAI’s math results: three withdrawals and criticism from the field. OpenAI has updated the GitHub repository featured in an earlier roundup — which contains hundreds of math manuscripts written by an unreleased model — with six new formal proofs, 19 modifications and three withdrawals. According to OpenAI researcher Dan Roberts, this means about 42% of the top-line results have now been verified in Lean, a proof-checking language. According to TechCrunch, the company did not follow the recommendations of the mathematicians’ advisory group convened by the Institute for Advanced Study in Princeton: of more than 700 manuscripts, the model’s chain of thought was released for only ten, and the group’s very first request — not to test open math problems on closed models — was not followed at all. According to mathematician Terence Tao, the problems are being “solved” by people who do not understand the output well enough to engage meaningfully with the field about it. Source: Dan Roberts (X), TechCrunch.
OpenAI: a Russian influence operation used ChatGPT — and also deceived schools and media outlets. OpenAI has taken down two AI-powered influence operations that spread geopolitical messages using fictitious journalists and a fake think tank. According to the report, previewed by NBC News, one of them, a Russian operation, deceived Latin American schools, a politician and media outlets, and stoked tensions between Ukraine and Poland; OpenAI says it is the most wide-reaching influence operation it has taken down so far. Source: OpenAI, NBC News.
USA Today’s publisher has also sued OpenAI. USA Today Co. and several of its local papers (including The Tennessean, IndyStar, The Columbus Dispatch and The Oklahoman) allege that OpenAI copied “hundreds of thousands” of their articles without permission to train its models, and are seeking more than $250 million in damages. It is the latest in a series of copyright lawsuits: OpenAI has also been sued by The New York Times, Ziff Davis, Encyclopaedia Britannica and a coalition of nearly 400 local newspapers, among others. Source: The Verge.
SpaceXAI: Grok Bot can now search X and manage Shopify stores. The official Grok Bot account announced late on Wednesday that the bot can now search, read and monitor X; on Thursday came the Shopify integration, which lets the bot check orders, track inventory and keep product descriptions up to date. Video 1.5 Lite, which creates video from text and images, became available in the Grok Imagine API ($0.02 per second at 480p, $0.03 at 720p and $0.14 at 1080p). SpaceXAI also became a founding supporter of the foundation behind the Omarchy Linux distribution with $1.5 million worth of Grok tokens; The Verge notes that the project’s founder, David Heinemeier Hansson, is controversial because of his anti-immigration writings. Source: Grok Bot (X), Grok Bot (X), Grok Imagine (X), The Verge.
Money for AI: a $3.1 billion valuation for Arena, more than $500 million for Manus. Arena, which ranks AI models by crowdsourced voting (it started as a research project at Berkeley), closed a $200 million Series B at a $3.1 billion valuation — up from $1.7 billion in January — led by Lightspeed and Khosla Ventures. It also unveiled the Alignment Index, which uses more than 90,000 real agent sessions across 27 models to measure things such as how often agents take actions they were not authorized to take. Butterfly Effect, the parent company of Chinese-founded Manus, raised more than $500 million; it is its first round since Meta’s $2 billion acquisition had to be unwound on the orders of Chinese authorities. The round was led by Boyu Capital and IDG Capital, with Tencent also participating. Source: TechCrunch (Arena), Arena (X), TechCrunch (Manus).
What this means for you: Google, Anthropic and OpenAI are all offering increasingly autonomous agents and workspaces, and the Gemini agent can already use rival Claude models — so choosing a model is becoming less and less a decision tied to a single vendor. If you want to roll out agents in your team, see how we build them. The confusion around OpenAI’s revenue figures and the tightening usage policies also show that it is worth planning for multiple providers and keeping an eye on the terms; our vendor-neutral consulting can help with this.