Breaking

OpenAI Patches ChatGPT macOS Flaw That Could Have Exposed Stored Chats

October 2, 2026 at 08:34 EDT

Security
AI Agents

OpenAI has patched a vulnerability in the macOS ChatGPT desktop app that researchers said could have let malicious local software take over the app and reach stored chats, browser sessions and other sensitive data. The fix appeared in OpenAI's September 25 macOS security update 26.924, which lists CVE-2026-100754 as fixed in version 26.924.20706 and credits Patrick Wardle of the Objective-See Foundation.

Oct. 2, 2026 · OpenAI

ChatGPT Mac flaw patched

A macOS vulnerability could have let local malicious code take over the ChatGPT app path and reach stored chats, browser sessions and connected data. OpenAI fixed CVE-2026-100754 in build 26.924.20706.

Fixed build

26.924.20706

Patch date

Sep. 25

Process checks

3 layers

Reported PoC

~12 lines

Why the small proof mattered

~12 code linesreported proof of concept

The issue was not measured by a benchmark or price tag. Its weight came from leverage: a compact local exploit path could route requests through software the user had already trusted.

Reported attack path

Local code

Attacker first gets code or a command running on the Mac.

→

Trusted interpreter

A trusted component accepts an untrusted script.

→

Main app process

The request appears to come through trusted software.

→

Stored data

Chats, browser sessions and connected app context are at risk.

What lowers risk

A fixed build is available, and the described scenario requires local code or commands first.

What raises stakes

Desktop AI tools can sit close to chats, files, browser context and connected services.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.