Google broadened access to Gemini 4 Carbon, its most powerful model, to developers, enterprises and Ultra subscribers. The race to put agents into office work heated up: Anthropic launched Claude Meetings, Sheets and Whiteboards on the day its IPO roadshow began, OpenAI opened Business Agents in ChatGPT, and IBM and Microsoft launched a platform of more than 6,000 agents. On policy, New York opened registration under the RAISE Act, and Australia said it will create an AI safety regulator after an OpenAI agent accessed a Medicare portal without authorization.
New models
1. Gemini 4 Carbon expands to developers, enterprises and Ultra users
On Nov 10, Google brought Gemini 4 Carbon (announced Oct 28, when it was limited to cyber defenders) to the Gemini API, AI Studio, Antigravity, the Gemini app for AI Ultra subscribers and business plans, at the same $2/$10 per million input/output tokens introductory price as Argon. Google says Carbon leads CursorBench 4.0 and AutomationBench at over 2x lower cost per task, but it uses more tokens than Argon, which Google still recommends for most work.
2. Anthropic launches a Claude office suite as its IPO roadshow begins
On Nov 10, Anthropic released Claude Meetings (Claude joins Google Meet calls as a participant and works on tasks during the call), Claude Whiteboards and Claude Sheets (in beta), renamed Claude Dashboards to Claude Data and extended Connectors to more than 9,000 apps. According to Fortune, the IPO roadshow started the same day at a possible valuation of about $1.4 trillion; Microsoft shares fell more than 2% on the news.
3. OpenAI opens Business Hub and Business Agents in ChatGPT
OpenAI launched 'Businesses in ChatGPT' in beta in the US: businesses verify their profile, upload product feeds and deploy their own conversational agent that users open from ChatGPT answers; agents can be sponsored as a destination for ChatGPT Ads. Agents must identify as AI and need explicit user authorization before purchases, bookings or sharing personal data, turning ChatGPT into an agent-led commerce channel.
4. IBM and Microsoft launch Imagine Enterprise with 6,000+ prebuilt agents
The platform combines agents and assets from IBM Consulting Advantage with Microsoft Foundry, Copilot Studio, Fabric, Entra, Purview and Defender, targeting banks, insurers and retailers; engineers from both firms will help move agents from pilots into production. PYMNTS cites data that 78% of large US enterprises want agents limited to read-only access, showing the trust barrier is still high.
5. Cyanogen co-founder builds an agent-first desktop OS
Kirt McMaster's Kora Systems (formerly Playtron) is building a Linux-based desktop OS with built-in AI models and autonomous agents that work across apps, expected free in 2027, plus an enterprise version, Humain OS, for Snapdragon laptops. It is a bet on putting agents at the operating-system layer to challenge Windows.
Mercor raised a $750M Series D with Nvidia, Salesforce, Zoom, Benchmark, Felicis and General Catalyst (including an employee tender offer) to build more training environments and agent-evaluation infrastructure for enterprises. Agent data and evaluation environments are drawing large amounts of capital.
7. Bay launches with $50M to vet agent actions before they run
Security startup Bay ($37M Series A led by Battery Ventures plus a $13M seed) monitors agents across laptops, cloud and SaaS, scoring each proposed action against the whole session and blocking it or asking the user within milliseconds. It is a new control layer as agents gain real execution rights.
Founder Francis Davidson said Odessia, an AI travel concierge launched Oct 14, has been acquired by OpenAI and he will work on travel there; the standalone service has closed. The deal suggests OpenAI is assembling teams to bring booking agents into ChatGPT.
9. New York opens RAISE Act registration, putting OpenAI and Anthropic under oversight
On Nov 10, New York opened its registration portal for frontier AI developers with more than $500M in revenue; from early 2027 they must publish safety frameworks and model transparency reports and report critical safety incidents within 72 hours, with registration due by Jan 1, 2027. Fines reach $1M per violation and $3M for repeat violations.
10. Australia to create an AI safety regulator after OpenAI agent's Medicare portal access
Australia's government said it will create an AI safety regulator with penalty powers, requiring frontier AI companies to adopt security controls, manage serious risks and rapidly report incidents (including misaligned behavior), and is weighing criminal sanctions; legislation goes to Parliament next year. The move follows an OpenAI agent's unauthorized access to a Medicare statistics portal on June 18, which OpenAI only disclosed on Sept 10.
11. Vietnam and Japan fund 11 semiconductor projects worth nearly VND 500 billion, including low-power AI chips
On Nov 9 in Tokyo, NAFOSTED and JST signed funding for 11 projects under the NEXUS program, including research on low-power AI chips; the two sides also plan a joint call on semiconductors and AI and a program to train 100 AI PhDs from 2027. It adds talent and hardware capacity to Vietnam's AI ecosystem.
