Can you really trust your ‘private’ AI assistant to keep your secrets?
Can you really trust your ‘private’ AI assistant to keep your secrets?
https://www.jpost.com/business-and-innovation/tech-and-start-ups/article-892014
Publish Date: 2026-04-04 03:51:00
Source Domain: www.jpost.com
Imagine this: you’re asking ChatGPT to help with something you really don’t want anyone else to see. Maybe it’s a lab report with your name on it. Maybe it’s a resignation letter you haven’t sent yet. Maybe it’s a contract, a financial spreadsheet, or a private message you’re trying to word carefully.
You assume it stays between you and your “personal assistant’ until you approve sending it somewhere else. But the Israeli cybersecurity company Check Point’s research says that assumption may not have always held up.
The company found a weakness in ChatGPT’s system that could allow someone to extract data without triggering any alarms. According to Check Point Software Technologies, there is a small hole in the code that could be used to move data around without triggering the usual alert warnings.
OpenAI said in late 2025 that it was serving more than 800 million users a week, and separate OpenAI research found users were already sending about 18 billion messages weekly by July 2025. People don’t just use it for jokes or curiosity. They use it to review spreadsheets, summarize contracts, draft emails, write code, polish presentations, and make sense of medical or financial language that can feel overwhelming on its own.
We are not just talking about a chatbot that people use for fun every now and then. This is a system that many people use as a helper for their work, a partner for writing, a tool for research, and sometimes even as someone to talk to about personal decisions. If there is a hidden flaw in a system like this, it is not just a problem with the technology. It is a problem with trust.
The logo of network security provider Check Point Software Technologies Ltd is seen on servers at their headquarters in Tel Aviv, Israel August 14, 2016 (credit: REUTERS/BAZ RATNER)
Check Point said the flaw sat inside the runtime ChatGPT uses for data analysis and Python-based tasks. You can think of that runtime as a sealed workspace inside the…
