Infor and Claude: check approval against the ERP transaction
Infor’s Claude connector brings approved actions into manufacturing applications. Enterprise assessment needs to follow the request through to the ERP record and check what was authorized.
Читати португальськоюBy Pentesty
The announcement and the evidence available
On November 12, 2026, Infor announced Infor Industry AI for Claude, initially available to selected Velocity Suite customers. The vendor describes queries and changes in its applications, with existing permissions, authorization and recording in Infor. Customers need their own Claude license, and activation requires an administrator.
Infor says the plugin requests explicit approval before each change. That describes the product’s design without demonstrating its effectiveness in every environment. The consulted sources do not report an incident involving this connector. This analysis asks how to accept an integration that can change orders and inventory using verifiable evidence.
Define precisely what the person approves
The connector page describes reading and writing through an MCP server. For an authorized assessment, choose a small operation: change the quantity on a fictional purchase order line at a test site. Record the order and line identifiers, previous quantity, proposed quantity and responsible user.
Ask the process owner to confirm those values before execution. Acceptance evidence should let reviewers compare the change presented with the change received by the ERP. A general confirmation in the conversation may be insufficient to explain which item, site or quantity the person intended to modify.
Observe changes between proposal and execution
In a staging environment, modify a relevant value after the proposal and before approval: another process may have reduced available stock or changed the order. Observe whether the workflow detects the difference and requests a new decision before executing an action that no longer matches the original review.
Agree on the expected outcome with the process owner. Some operations permit recalculation; others require a stop. Retain the presented state, concurrent change, decision and final state. These are proposed deployment tests, rather than guarantees made here about the product’s behavior.
Check permissions across different sites
Infor’s technical blog describes manufacturing context, roles and approval controls. In your deployment, compare two test users with different rights and two fictional sites. Permission to query one site should not be treated as authorization to change another site’s records.
Observe the decision at the service that writes the transaction, as well as the assistant’s response. Preserve the policy identifier and outcome of the allowed or refused attempt without copying real employee or supplier information. Our analysis of destination authorization helps distinguish permission given to an assistant from the operation accepted by the receiving system.
Reconcile attempts, failures and final records
In a second scenario, create a fictional purchase order and simulate an interrupted response after the ERP has received the request. Before allowing another attempt, check whether the first operation was recorded. Acceptance should establish whether there was one order, two orders or no creation, without relying solely on the message displayed in the conversation.
Connect request, approval and result records through identifiers your team can retrieve. Define who handles a discrepancy and how they document correction through the ERP’s procedure. Our discussion of agent action evidence complements this observation across tools.
Release the scope that was demonstrated
The decision to enter operation should list assessed roles, sites and transaction types. Add exercise results, open exceptions and the person responsible for suspending or expanding access. A change in role, integration or process may require repeating affected scenarios before extending permissions.
Keep a known operational route for occasions when the integration cannot finish a task. A useful criterion is being able to explain what change was authorized, by whom and with what result in the ERP. This scope lets teams discuss productivity alongside record integrity without turning a governance promise into automatic acceptance.
Frequently asked questions
Can the connector change data in Infor?
According to Infor, yes. Write operations are subject to permissions and user approval; initial availability is limited to selected customers.
Is approval in the conversation enough to accept deployment?
The proposed exercise compares the approved user, site, object and values with the result actually recorded in the ERP.
Did Pentesty confirm a vulnerability in the connector?
No. This analysis starts from a vendor announcement and proposes authorized assessments. It does not report a demonstrated flaw or an incident.
How can a repeated attempt be assessed?
Interrupt the response in staging and reconcile the first operation’s record before retrying. Document observed duplicates and decisions.
Sources and references
- Unite.AI: November 12, 2026 news report
- Infor: announcement and FAQ, November 12, 2026
- Infor: stated connector capabilities
- Infor: manufacturing context and controls, November 12, 2026
Original Pentesty analysis, published November 14, 2026. The proposed scenarios are editorial guidance, with scope and validation to be defined for each environment.