ByDANIEL COHEN

JULY 22, 2026 15:20

Artificial intelligence continues to transform payroll practices, yet fresh survey data indicates that many accountants are applying the technology in ways that risk exposing confidential employee records and could breach Israeli privacy regulations.

A study by Oketz Systems found that 85 percent of payroll accountants sent at least one payslip to an AI platform in the past year for help with work questions. This change has reduced reliance on older methods. Previously, payroll staff mainly used specialized Facebook communities for guidance, but activity across four leading groups dropped by roughly 45 percent.

Unsecured AI platforms displace established payroll resources

Payroll teams now favor open AI services and upload payslips directly rather than using conventional forums or protected company systems. Secure enterprise AI solutions avoid training on client data and remain under organizational security protocols.

The impact of AI appears early in the payroll cycle. Each month, departments gather attendance logs, commission figures, vehicle valuations, expense claims, and other details from clients via spreadsheets, messages, and assorted programs. Until recently, staff entered every item by hand.

With a clear prompt, AI can merge these inputs into one table ready for payroll software. The tools can also review anomalies on their own by scanning month-to-month patterns, spotting unusual shifts, and suggesting explanations, which removes the need for manual spreadsheet reviews.

Outcomes include fewer mistakes, steadier workloads across the month, and more time for accountants to focus on review and expert decisions instead of routine entry tasks.

A key difference exists in deployment methods. While company programs center on protected AI agents, the survey showed many accountants sent payslips to ordinary public AI services rather than secure enterprise options, creating notable security exposure.

Security and legal issues put employee salary data at risk

Sending payslips to services such as ChatGPT moves files beyond company networks to servers of overseas firms and their partners worldwide. On personal accounts, default options may allow content to train future models unless users turn that feature off.

After upload, organizations face several threats, including stolen login details that expose chat records, software flaws like the 2023 case that leaked chat titles and partial billing data, and the chance that models store or repeat personal information.

Details on wages, employers, deductions, and retirement plans could aid cybercriminals in targeted scams, a known tactic to divert salaries by altering bank records.

The approach may also breach Israeli statutes. Amendment 13 of the Protection of Privacy Law treats payroll information as highly sensitive data needing stronger safeguards. Section 16 bars sharing data outside work needs and allows up to five years imprisonment, while Section 17 places personal duty on protecting records.

Moreover, the mandatory prior contract with an external processor under Regulation 15 of Israel's Information Security Regulations is typically missing, a standard consumer AI providers fail to satisfy. After 2025 guidance from the Privacy Protection Authority requiring informed consent for AI processing of personal data, the clear takeaway is that AI can assist work but payslip details should stay out of public tools.

Call for targeted training and custom solutions

"The AI shift in payroll accounting has arrived," stated Haim Molcho, CEO of Oketz Systems. "It represents more than an added option. It marks a core change in daily operations. Yet realizing its benefits while staying within privacy rules requires focused training, which is why we launched a dedicated program."

"The exposure comes from two directions," he added. "Legal risk arises from improper AI practices, and errors occur when generic tools lack proper adaptation. Payroll programs already include labor rules, but each office follows unique steps for client files, spreadsheet layouts, import formats, and checks. A tool untrained on those steps produces results that do not fit payroll systems, and issues may surface only after the fact."