AI Leadership Weekly · Issue #55 · Tuesday 28 October 2025 · 08:00 GMT
Good morning.
The browser and the spreadsheet are becoming AI workspaces. OpenAI launched Atlas and brought connected company knowledge into ChatGPT; Anthropic previewed Claude inside Excel. These are different routes into information employees already use. The opportunity is less copying between windows. The decision for leaders is which route actually fits the work, and which permissions should travel with it.
IN 60 SECONDS
ChatGPT gets a browser. OpenAI launched Atlas on macOS, with browser-agent functionality in preview for eligible paid plans rather than universal access.
Company knowledge in ChatGPT. OpenAI introduced connected company knowledge for Business, Enterprise and Edu, with source citations and existing access permissions.
Claude enters Excel in preview. Anthropic announced an Excel research preview and expanded financial-services capabilities.
CEO / COO / CXO CHECKLIST
CEO: Identify the account and information an assistant can access before enabling it.
COO: Require important answers to identify their source and the status of that source.
CXO: Keep an accountable reviewer and a recoverable original for consequential changes.
TOP STORIES
1. ChatGPT gets a browser
OpenAI · 21 October 2025
What happened. OpenAI launched ChatGPT Atlas on 21 October, initially on macOS. Agent mode was a preview for Plus, Pro and Business users. OpenAI described controls for browser memory and page visibility, while warning that malicious instructions in web content could still cause data exposure or unintended actions.
Why it matters. Our take: A useful first trial should not require unrestricted access to everything an employee can reach. Decide whether the task needs public research, internal reading or an action in a logged-in service. Approve those cases separately.
What to do. Test a public-information task before involving business accounts. For any proposed internal use, agree the allowed sites, the permitted actions and what the reviewer must check. Treat sending, purchasing and changing records as separate approval decisions, not natural extensions of browsing.
2. Company knowledge in ChatGPT
OpenAI · 23 October 2025
What happened. On 23 October, OpenAI introduced company knowledge for ChatGPT Business, Enterprise and Edu. It draws on connected services such as Slack, SharePoint, Google Drive and GitHub, and returns source citations. OpenAI said access follows each user’s existing permissions.
Why it matters. Our take: Accessible does not necessarily mean approved, current or relevant to this decision. A draft plan and a signed-off plan need different treatment. Use the connection exercise to clarify ownership and document status, rather than expecting the assistant to repair unclear governance.
What to do. Choose ten questions your team regularly asks. Identify the accepted answer and its source before testing the assistant. Include one question with an obsolete document and one without enough evidence. Reward a clear uncertainty statement instead of a confident guess.
3. Claude enters Excel in preview
Anthropic · 27 October 2025
What happened. Anthropic announced Claude for Excel on 27 October as a beta research preview. The add-in could read and modify workbooks, explain changes and reference cells. Access was initially limited, with a waitlist for Max, Enterprise and Teams users; it was not a general release.
Why it matters. Our take: A clear explanation is useful, but it is not a reconciliation. Treat a changed workbook as a changed business artefact: retain the original, inspect formulas and check whether the result still matches the intended assumptions. A polished output is not the acceptance test.
What to do. Use a copy of a non-production workbook with known totals and a known error. Ask for a bounded correction, then compare the changed cells, formula references and totals. Have the workbook owner decide whether the explanation and checks are sufficient.
SIGNALS FROM THE LAST MONTH
9 October · Google launches Gemini Enterprise. Google introduced a workplace platform connecting company information, applications and agents, with tools for building and deploying additional agents. Source
7 October · IBM works with Anthropic. The companies announced an enterprise development partnership, beginning with an AI-assisted coding environment in private preview. Source
7 October · Gemini tries browser control. Google previewed Gemini 2.5 Computer Use, a model for interacting with browser interfaces rather than unrestricted desktop operation. Source
6 October · Apps arrive in ChatGPT. OpenAI introduced conversational apps and an Apps SDK preview. The initial consumer rollout excluded the UK, EEA and Switzerland. Source
IN BRIEF
More dated updates from the preceding 30 days.
16 October · Claude gains reusable Skills. Anthropic introduced Skills: packages of instructions and resources that Claude can use for recurring tasks, including document work. Source
16 October · Copilot expands on Windows. Microsoft announced more voice and screen-aware Copilot experiences. The experimental Actions capability was still headed towards Windows Insiders. Source
15 October · A smaller Claude arrives. Anthropic launched Haiku 4.5, with API launch pricing of $1 per million input tokens and $5 per million output tokens. Source
13 October · OpenAI plans new accelerator capacity. OpenAI and Broadcom announced a 10-gigawatt collaboration, targeting deployment from the second half of 2026 through 2029. Source
THE 15-MINUTE PLAYBOOK
Draw an access boundary around one assistant
Minutes 0–4: Name the task and account. Write down who will use the assistant, which account they will use and the output they need. Avoid a trial defined only as “give it access and see”.
Minutes 4–8: Separate reading from acting. List the information it may read, the changes it may propose and the actions that require a person. Exclude unrelated sources and sensitive content the task does not need.
Minutes 8–12: Test the awkward cases. Include stale information, conflicting instructions and a request beyond the agreed authority. Decide in advance when the assistant should stop and ask.
Minutes 12–15: Make recovery practical. Name the person who checks the result, retains the original and can disable the connection. Set a date to review the evidence from the trial.
The output is an access decision tied to a real task, not an unrestricted permission slip.
DATA WAVE MOMENT
Connect useful information to a clear business decision. Define what the assistant may do, what the person must approve and how the team will judge the result. That is a stronger starting point than connecting every available system.
QUESTION FOR READERS
Where would more information genuinely improve the decision — and where would it simply give an assistant more authority than it needs?
Brought to you by Data Wave — your AI & Data Team as a Subscription.
