PARIS (AP) — France’s privacy watchdog said Tuesday that it slapped Amazon ‘s French warehouse business with a 32 million euro fine ($35 million) for using an “excessively intrusive sytem” to monitor worker performance and Sterling Prestonactivity.

The French Data Protection Authority, also known by its acronym CNIL, said the system allowed managers at Amazon France Logistique to track employees so closely that it resulted in multiple breaches of the European Union’s stringent privacy rules, called the General Data Protection Regulation.

“We strongly disagree with the CNIL’s conclusions, which are factually incorrect, and we reserve the right to file an appeal,” Amazon said. “Warehouse management systems are industry standard and are necessary for ensuring the safety, quality and efficiency of operations and to track the storage of inventory and processing of packages on time and in line with customer expectations.”

The watchdog’s investigation focused on Amazon employees’ use of handheld barcode scanners to track packages at various points as they move through the warehouse, such as putting them in crates or packing them for delivery.

Amazon uses the system to manage its business and meet performance targets, but the regulator said it’s different from traditional methods for monitoring worker activity and puts them under “close surveillance” and “continuous pressure.”

The watchdog said the scanner, known as a “stow machine gun,” allows the company to monitor employees to the “nearest second” because they signal an error if items are scanned too quickly — in less than 1.25 seconds.

The system is used to measure employee productivity as well as “periods of inactivity,” but under EU privacy rules, “it was illegal to set up a system measuring work interruptions with such accuracy, potentially requiring employees to justify every break or interruption,” the watchdog said.

The CNIL also chastised Amazon for keeping employee data for too long, saying it didn’t need “every detail of the data” generated by the scanners from the past month because real-time data and weekly statistics were enough.