ChatGPT Just Moved Into Your Oracle Apps. Who Controls The Data?

Today, we're diving into:

ChatGPT Just Moved Into Your Oracle Apps. Who Controls The Data?

At Oracle AI World in Las Vegas on Tuesday, Oracle and OpenAI announced a deal that puts ChatGPT, Codex and OpenAI's models inside the systems that run your finance, supply chain and operations, according to Oracle. Through Oracle AI Workspace, plugins in ChatGPT let employees pull information, complete approved business workflows and take action in Oracle systems.

That list of systems is the part to read twice: Oracle Fusion Applications, NetSuite, Oracle AI Database and Oracle Cloud Infrastructure. The chat window is no longer next to your system of record. It is wired into it.

Your AI Can Now Act In Your ERP. Does It Know What Your Data Means?

Find out which of your business definitions, owners and permissions an AI agent would trip over first.

What Oracle And OpenAI Actually Announced

The announcement covers four pieces. Each one moves AI a step closer to where real business decisions are made.

Oracle says apps built in App Composer connect only to approved Oracle data and workflows, and inherit enterprise identity, permissions, networking, audit and lifecycle controls. That is the right design instinct.

What The Press Release Does Not Say

The release gives no general availability dates for most of these products, and it carries the usual disclaimer that timing and pricing may change. It also does not explain data residency, how customer data is treated for model training, or the specific access-control mechanisms behind the plugins.

The Real Bottleneck Is Not The Plugin. It Is Your Permissions.

Here is what changes. When AI only answered questions, a messy permission model produced an awkward answer. When AI completes workflows, the same messy permission model produces an approved purchase order, a changed vendor record or a posted journal entry.

"Inherits your permissions" sounds reassuring. But inheriting a permission model built for humans who log in twice a day is very different from inheriting it for an agent that acts hundreds of times an hour.

The wrong reaction is to block ChatGPT from Oracle entirely and wait. Your teams will find another route, and you will lose visibility. The better alternative is to treat this as an access review with a deadline.

Building that layer of meaning and ownership starts with trusted enterprise context. DataManagement.AI helps organizations create that foundation by unifying metadata, lineage, governance, business glossaries and knowledge assets into a single intelligence layer. It automatically discovers metadata, maps lineage across the modern data stack and identifies downstream dependencies, so before an AI agent touches a record, you know what that record means, where it flows and who owns it.

Apple Just Opened Your Documents To AI Agents. Who Gets To Edit Them?

On Tuesday, Apple announced that Keynote, Pages and Numbers now support Model Context Protocol (MCP) on macOS 27, iThinkDiff reports. With MCP switched on, outside AI agents can read document content, make edits and run automated actions directly inside all three apps.

Apple also introduced the Keynote Interchange Format (.keyx), a JSON-based file built to be readable by external tools, developer workflows and AI agents. And the three apps get built-in assistants that run on Apple's own models and on OpenAI's.

Why This Matters More Than It Looks

MCP is the open standard Anthropic introduced for connecting AI models to apps and data. Think of it as a universal adapter: any agent that speaks MCP can plug into any app that does.

Until now, most enterprise MCP conversations were about databases, ticketing systems and code repositories. Apple just brought it to the board deck, the contract draft and the budget spreadsheet sitting on an executive's laptop.

The Fine Print For Your Rollout

This Is Not A Productivity Feature. It Is A New Data Perimeter.

Picture your CFO's laptop the week before earnings. A draft results deck in Keynote, a forecast model in Numbers, and a personal AI agent the CFO installed because it saves an hour a day. With MCP on, that agent can read every slide and every cell.

Nobody did anything wrong. But your most sensitive numbers just became reachable by software your security team has never reviewed.

The wrong reaction is to ban MCP on every Mac. The better alternative is to decide, in writing, which agents may connect to which apps, and on whose devices.

The Dutch Government Is Cutting 25,000 Jobs With AI. The Plan Comes Later.

The Netherlands plans to cut about 25,000 civil service positions over several years as part of an AI-led overhaul, Bloomberg reports via The Star. State Secretary for Effective Government Eric van den Burg set out the plan in a letter to parliament, aiming for "simpler, smarter, and smaller" government.

The savings target is €900 million a year by 2027, rising to €3.5 billion a year from 2035. The detailed plan for how AI will raise productivity is due early next year.

The Numbers Behind The Decision

Central government headcount hit a record 160,000 at the end of 2025, up 41% from 2018, despite earlier promises to shrink it. Ministries, inspectorates and implementing agencies will be hit hardest, with some departments facing cuts of 45% or even 60%. Police and defense are exempt.

It Is Not Only AI

Van den Burg pointed to non-binding parliamentary motions calling for a 20% productivity gain through AI. But the plan also leans on fewer external hires and a program to scrap 500 unnecessary rules a year. The government wants to rely mainly on Dutch and European suppliers to avoid depending on non-European technology.

The Real Risk Is Cutting Before You Can Measure

This is the pattern to watch in your own boardroom. A headcount target arrives first. The evidence for how AI will absorb the work arrives months later.

This is not a technology problem. It is a sequencing problem. If you remove the people who know how a process really works before AI has proven it can run that process, you do not get savings. You get backlogs, errors and expensive rehiring.

Know Your Inference (KYI) is where this comes together. When you understand what each AI workload costs to run, how fast it performs and how accurate it is on your real cases, you can set headcount and savings targets on evidence, instead of setting them first and hoping the technology catches up.

Your opinion matters!

Hope you loved reading our piece of newsletter as much as we had fun writing it.

Share your experience and feedback with us below 'cause we take your critique very critically.

Thank you for reading

-Shen & Towards AGI team