Artificial intelligence is rapidly reshaping the payroll profession, but a new survey suggests many payroll accountants are using the technology in ways that expose highly sensitive employee information and may violate Israeli privacy laws.
According to a survey conducted by Oketz Systems, 85% of payroll accountants uploaded at least one payslip to an AI system over the past year to receive assistance with a professional issue. The shift has come largely at the expense of traditional consultation channels. Before the rise of AI, payroll professionals primarily turned to dedicated Facebook groups for advice, but the company's review of four major professional groups found activity had fallen by approximately 45%.
<strong>Unsecured AI tools replace traditional payroll channels</strong><br>
Payroll professionals are increasingly consulting unsecured AI tools and uploading payslips to them instead of relying on traditional consultation platforms or protected enterprise systems. Enterprise AI platforms do not use customer data to train models and operate under an organization's security controls.
The AI revolution in payroll accounting is most evident even before payroll calculations begin. At the start of every month, payroll departments receive attendance records, commission reports, company car valuations, expense reimbursements, and other information from clients in Excel files, emails, and various software systems. Until recently, every piece of data had to be collected and entered manually.
Today, a well-crafted prompt allows AI tools to consolidate all those sources into a single table that can be imported directly into payroll software. The systems can also investigate irregularities independently by comparing data across months, identifying unusual increases or decreases, and flagging their likely causes, eliminating the need for payroll accountants to manually comb through spreadsheets.
The result is fewer errors, a workload spread more evenly throughout the month instead of piling up during payslip production, and payroll professionals who can devote more time to oversight and professional judgment rather than repetitive data entry.
However, there is a critical distinction in how these tools are being deployed. While organizational efficiency initiatives are built around dedicated, secure AI agents, the survey found that many payroll accountants uploaded payslips over the past year to standard public AI platforms rather than secure enterprise systems, a practice that creates significant security risks.
<strong>Cybersecurity, legal risks threaten sensitive employee data</strong><br>
Uploading payslips to AI systems such as ChatGPT transfers documents out of an organization's internal environment and onto the servers of foreign commercial companies and their subcontractors around the world. On personal accounts, the default settings may also permit submitted content to be used for future model training unless users explicitly disable that option.
Once the information has been uploaded, organizations become exposed to several potential risks, from compromised login credentials that reveal chat histories to vendor software bugs, such as the 2023 incident in which chat titles and partial billing information were exposed. There is also the possibility that models could retain and reproduce personal training data.
Sensitive information relating to salaries, employers, wage garnishments, and pension funds could become a valuable tool for cybercriminals conducting targeted phishing attacks, a documented method used to steal salaries by fraudulently changing employees' bank account details.
The practice may also constitute a clear violation of Israeli law. Amendment 13 to the Protection of Privacy Law classifies payroll data as particularly sensitive information requiring enhanced security measures. Section 16 prohibits the disclosure of information unrelated to work purposes and carries penalties of up to five years in prison, while Section 17 imposes personal responsibility for safeguarding information.
In addition, the required prior agreement with an external data processor under Regulation 15 of Israel's Information Security Regulations is generally absent, a condition that consumer AI providers do not meet. Following guidance issued by Israel's Privacy Protection Authority in 2025, which requires informed consent for processing personal information using artificial intelligence, the practical conclusion is straightforward: AI tools may be used, but personally identifiable information from payslips should not be entered into them.
<strong>The need for dedicated training and specialized tools</strong><br>
"The AI revolution in payroll accounting is already here," said Haim Molcho, CEO of Oketz Systems. "This is no longer just another feature. It is a fundamental change in the way payroll professionals work. But anyone who wants to realize its potential without running afoul of privacy laws must receive dedicated training and guidance, which is exactly why we established a specialized course on the subject."
"The risk is twofold," he continued. "There is legal exposure resulting from unlawful AI use, and there is also the risk of faulty work when generic AI tools are used without proper customization. Labor regulations and legal requirements are already built into payroll software, but every office has its own unique workflow: The client files it receives, the structure of its spreadsheets, the import formats, and its verification procedures. A tool that has not been trained on those workflows will generate outputs that cannot be properly integrated into payroll systems, and the accountant may discover the problem far too late."