Artificial intelligence has become a silent threat in the office: The biggest data leak is on ChatGPT

A recent study by Harmonic Security based on 2025 data shows that the artificial intelligence platforms used in firms pose a much greater security risk than previously thought. The report states that although OpenAI's ChatGPT service has a share of less than 50 percent in corporate AI usage, more than 71 percent of data leaks occur through this platform.

The analysis examined a total of 22.4 million AI requests, and the majority of sensitive data was exposed through just six popular applications. These applications stand out as the source of 92.6 percent of the leakage risk.

SOFTWARE CODE AND LEGAL DOCUMENTS ARE LEAKED THE MOST

These leaks, which jeopardize company information, particularly include confidential software code and internal legal documents. It was determined that approximately 30 percent of the 579,000 sensitive data entries consisted of source code from the software sector, 22.3 percent were legal documents that should remain confidential, and 12.6 percent were data on company mergers or acquisitions. Employees often sharing technical or strategic information about business processes directly on AI platforms under the assumption that it is "unimportant" further increases security vulnerabilities.

FREE ACCOUNTS BROKE THE FIREWALL

Another striking finding of the report is that data sharing occurs largely through free accounts. Most sensitive data is entered into the system through "ChatGPT Free" accounts opened by employees themselves, and this information can often be used to train public models. Only 13 percent occurs in corporate accounts that can be controlled by company policies.

In addition, it is noted that other major AI platforms such as Microsoft Copilot and Google Gemini carry similar risks; however, the main weight of the leaks is ChatGPT-sourced. Furthermore, while the report reflects that approximately 4 percent of corporate requests are made on China-based applications such as DeepSeek, this situation increases international data security concerns.

WARNING FROM EXPERTS: CONTROLLED USE INSTEAD OF BANNING

Security experts believe that the problem cannot be solved simply by banning artificial intelligence tools. According to them, blocking access to these platforms could drive employees toward more uncontrolled and harder-to-detect methods. Instead, it is recommended that companies invest in intelligent security infrastructures capable of monitoring data flow in real-time and provide their staff with artificial intelligence literacy training. Experts warn, "Every detail shared on artificial intelligence can expose an organization's most intimate secrets."

News Source: 12punto