Today's thread
This week AI companies moved beyond the chat box into work tools. Anthropic added spreadsheets and whiteboards inside Claude, and Google opened its strongest model, Gemini 4 Carbon, together with a feature that lets several AIs work as a team. OpenAI made ChatGPT Enterprise purchasable on the AWS bill companies already pay.
The deeper these tools go into work, the heavier the rules and security load becomes. Anthropic's new usage policy takes effect today (November 12), and a reported leak of about 100 Korean business Claude accounts made the news. Sorting out permissions and internal rules now matters as much as adding features.
Claude gets Sheets and Whiteboards; Claude Data goes GA
On November 10 Anthropic launched collaboration tools that live inside Claude. Claude Whiteboards is a shared board where a team brainstorms with Claude; describe a plan in chat and it becomes a board with owners and dates. Claude Sheets builds and edits spreadsheets with tabs, formulas, formatting and dropdowns (pick-from lists). Both are in beta (a pre-release trial) and available on all paid plans on web, desktop and mobile.
Claude Dashboards, launched in beta in October, has been renamed Claude Data and is now generally available. Connect a company data warehouse (for example BigQuery or Amazon Redshift) or a customer management system (CRM), and you can pull numbers with a one-line question. These links are called connectors, and the same day Anthropic expanded the ability to open partner apps such as Adobe, Atlassian, Box and Notion beside the chat or full screen and edit them with Claude.
In practice, tasks that used to bounce between Excel and a whiteboard tool, such as meeting notes, simple tallies and weekly metrics, can be finished in one conversation. Anthropic itself says Sheets offers simple spreadsheet functionality for now. For files that need complex functions, export to .xlsx or CSV and finish in Excel.
On Team plans all three are on by default. On Enterprise, Whiteboards and Sheets are off by default, and Claude Data turns on by default on November 19. Admins should decide before then in organization settings and list which data to connect. How to split access rights per dataset has to be designed by each company.
Gemini 4 Carbon opens to Ultra, with /swarm in preview
On November 10 Google opened Gemini 4 Carbon to Google AI Ultra subscribers. It is available in the Antigravity app, command-line tool (CLI) and code editor extensions, and businesses can use it through Gemini for business. The same day /swarm launched in preview: a coordinator AI splits the work across several worker AIs running in parallel, and verifier AIs run tests and challenge the results.
When announced on October 28, Carbon was first opened only to a subset of cyber defenders. Google calls it its strongest model and self-reports 88.2% on a vulnerability-finding evaluation (SEC-bench Pro). The introductory price on the API (the channel programs use to call the model) is $2 input and $10 output per million tokens, rising to $4 and $20 afterward. Tokens are the small chunks an AI counts text in.
/swarm lets teams try out a group of AIs that split one instruction into roles. Each role can use a different model, so costs can be divided, with the expensive Carbon on design and verification and cheaper models on repetitive work. Teams planning to adopt agents (AIs that carry out multi-step tasks on their own) can measure the effect on one small task.
Google itself warns that running many AIs in parallel uses up quota much faster than a standard chat. How Ultra subscription limits translate into API pricing has not been disclosed, and the Antigravity post names only Ultra subscribers as eligible. /swarm is in the app and CLI only, not yet in editor extensions.
ChatGPT Enterprise now buyable against AWS commitments
On November 9 OpenAI listed ChatGPT Enterprise on AWS Marketplace (Amazon's cloud software store). OpenAI is the seller, and there is no public price; each company gets a custom quote. The AWS listing says this spend counts toward a company's AWS commitment (EDP or PPA), subject to its agreement.
Large companies sign multi-year spending commitments with AWS in exchange for discounts. Until now ChatGPT Enterprise required a separate contract with OpenAI; now it can be bought within existing cloud budgets and procurement processes. OpenAI said eligible customers can also buy its healthcare and financial services editions of ChatGPT through the same route.
Companies whose AI adoption has been stuck at vendor onboarding and purchase approval now have a way to evaluate it under their existing AWS contract. Start by checking the remaining AWS commitment and renewal date with procurement and finance. Buying inside a cloud you already use may also cut security review paperwork.
Whether the Marketplace price matches a direct contract, and in which regions data is processed, has not been disclosed. Whether it applies as-is to contracts with Korean entities should be confirmed with an AWS or OpenAI account team.
Anthropic's new usage policy takes effect today
Anthropic's new usage policy, announced on October 8, takes effect today (November 12). The usage policy is the rulebook for what Claude may and may not be used for. The most visible change is a new ban on sustained and needless abusive or cruel behavior toward the models. Anthropic says ordinary frustration, pushback, dark creative themes and model testing are not covered; only extreme cases are targeted.
Rules on fake accounts, fabricated news sites and influence operations are merged into one section banning deceptive campaigns and artificial activity, covering commercial as well as political activity. In law enforcement, Claude may not be used to decide or recommend who to investigate, arrest or charge. Hardware that takes autonomous physical actions now requires a qualified operator who can observe and stop it.
Marketing teams should read the deceptive campaign clause first. Mass-creating reviews, testimonials or social accounts to look like real people is banned even for commercial purposes. Companies using Claude for customer service or internal automation should copy this clause into their internal AI guidelines.
Anthropic says most changes clarify existing rules. The announcement does not address new rules on account sharing or resale. It also does not detail how violations are detected and enforced.
About 100 Korean business Claude accounts reported leaked
Korean media reported on November 9 that Anthropic notified Korea's Personal Information Protection Commission on October 31 of a leak involving about 100 Korean business Claude accounts. Exposed items include account details such as names, email addresses, account IDs, organization IDs and plan tiers. Anthropic said prompts (what users typed), outputs and payment information were not accessed.
An outside security firm is believed to have reached Anthropic's internal systems through a vulnerability in third-party business software. The Washington Post identified the firm as US security startup Cathedral. Anthropic said it blocked the affected systems the day it learned of the access, fixed the vulnerability with the vendor, and told some customers their data may have been exposed.
Even without chat contents, names, emails and organization details are good material for targeted phishing (scam emails aimed at specific people). Company Claude admins should check whether they received a notice from Anthropic and warn staff about login or payment emails impersonating Claude or Anthropic.
The commission is at the preliminary review stage and has not decided whether to open a formal investigation. Because Anthropic is a foreign company, data request procedures may lengthen the review. The list of affected companies and the exact scale have not been disclosed.
Try today
If you administer Claude Team or Enterprise, decide before November 19 in organization settings whether to enable Claude Sheets, Whiteboards and Data, and write down which data to connect and who can see it.
Add one line from Anthropic's new deceptive campaign clause to your internal AI guidelines: fake reviews and mass account creation are banned even for commercial purposes. Share it with the marketing team.
If your company uses Claude, check whether you received Anthropic's leak notice and send staff a warning about emails impersonating Claude or Anthropic.
If your company runs on AWS, check the remaining AWS commitment with procurement and ask whether a ChatGPT Enterprise quote can fit within it.