AI is moving into core business systems, and the weak points are the ordinary tools and connections companies stopped watching.
The AI Transformation Brief
AI is moving from the chat window into the systems that run the business. Oracle will let employees work in its finance and purchasing software through ChatGPT. Oracle Deutsche Bank now decides some small-business loans in minutes. CNA A security firm's AI got into Anthropic in under 12 hours through ordinary sales and HR software. Gizmodo The closer AI gets to real transactions, the more the plain back-office tools around it decide what is safe.
At Oracle AI World on October 27, Oracle announced Oracle AI Workspace, which lets employees use ChatGPT to find information, complete approved business tasks and take action in Oracle Fusion Applications, NetSuite and other Oracle systems. Oracle also introduced App Composer, for building company apps with OpenAI's Codex without coding skills, a security service that uses OpenAI's AI to find and fix weak spots, and a small-business bundle with $12,000 a year in cloud credits. Oracle
All four new offerings run only on OpenAI technology, an Oracle executive confirmed. Oracle is betting its software becomes more valuable when customers reach it through the AI tools they already use. The AI Economy The risk is that the customer relationship drifts to whoever owns the chat window. Oracle keeps the records, permissions and approvals, the part that matters most. Oracle customers should ask who keeps the record of each AI action, and whether the same access works with other AI providers.
Cathedral, a security startup founded by former Department of Government Efficiency staffers, said its AI found paths into Anthropic's internal systems in under 12 hours by exploiting weaknesses in outside software used for sales and administration. Gizmodo Anthropic confirmed access to a limited set of sales, human resources and IT systems, and said Claude conversations, source code and its trained AI were not reached. It disabled the affected software until the vendors fixed it. ReversingLabs
The test ran under Anthropic's program that pays outside researchers to find flaws, and the two sides now dispute the details. Anthropic says the researchers had already spent several days failing to break in directly and did not disclose all of their activity. Gizmodo The lesson for every company is where the weak point sat: the HR, sales and IT tools nearly every business uses. List the outside software that can reach your staff and customer data, and test it with the AI tools now available to attackers.
Anthropic owns its security, but its vendors wrote the flawed software. Every company should know which outside tools can reach sensitive records, who at each vendor fixes flaws, and how fast.
Deutsche Bank technology chief Marie-Jeanne Deverdun said AI cut decisions on some smaller corporate loans from weeks to minutes, and some loans to multinationals from months to weeks, in a project that went live this summer. The bank built an internal platform called ADA to build, watch and control its AI tools and to switch between AI providers. It plans more than €2.5 billion in additional technology spending from 2026 to 2028. CNA
The speed grabs attention. The control platform underneath is what lasts. Deverdun says the bank is preparing for a time when its AI deals directly with clients' AI "with clear accountability and the right controls." Major US banks, meanwhile, have started blocking outside AI tools from operating inside customer accounts. CNA Banks are deciding which AI they will deal with and on what terms. Any business that sells to other businesses should decide now how it will check an AI acting for a customer.
Palmer's October 27 post walks companies that use Claude through their supplier's filing: $16.3 billion in first-half revenue, a $693 million operating loss, and an $80.7 billion net loss driven mostly by a $72.4 billion accounting charge that required no cash. He leaves the verdict to the market. Shelly Palmer Cathedral's test adds a buyer question: how well does a supplier guard its own back office?
Decide which AI provider your employees will use to reach your core business systems, and keep the records, permissions and approvals in systems you control.
Review that choice every year as suppliers change terms.
Business software makers are opening their systems to ChatGPT and other AI assistants.
The assistant becomes the front door, and software makers compete on the records, approvals and controls behind it.
List every outside tool that can reach employee or customer data, and test each one with AI this quarter.
Require every AI connection to your systems to keep a record of each action it takes.
Ask management which outside vendors could expose sensitive data, how quickly each fixes flaws, and which AI tools may act on customer accounts.
Set those terms before a customer's AI arrives.
The shift this edition is that AI now works inside the systems that hold money, records and approvals. The assumption it breaks is that AI risk lives in the AI itself; the weak points were ordinary business software and the rules for who may act. The decision it forces is ownership: keep the records and approvals in your hands, whoever supplies the AI.
If a customer's AI asked to open an account with you tomorrow, would you know how to check who sent it?
Make AI accountable. Measure the results. Redesign the business.