Specifically:
The data entry part of the Excel model was not automated as recommended by the internal reviews.
After catastrophe struck, the Excel model was found to have several simple mistakes in its calculations which profoundly compromised its results.
Neither of these are problems of reckless deception.
JPMorgan's own internal review mandated that they automate the inputs of Excel spreadsheets, but this is not easy to do. Cells can link to values in other spreadsheets, but although this removes a step in the maintenance process, it also makes mistakes even harder to detect, and allows them to propagate seamlessly to spreadsheets further along in the "chain." This goes to the core problem that it is impossible to build and enforce basic business logic in Excel.
The second issue is that Excel models present only results; it hides the code and the data flows behind those results. This causes two problems in practice, 1) it disincentivizes people from looking at and validating the actual code the model is based on, and 2) even if you try, you can never get a comprehensive overview of the code. You must look at it on a cell-by-cell basis, and piece together in your own head what is being used where.