ChatGPT Mac flaw exposes risks in trusted AI apps

On October 2, 2026, WIRED reported that researchers at the Objective-See Foundation had found a vulnerability in OpenAI’s ChatGPT app for macOS. According to WIRED, untrusted code already running on a Mac could potentially pass commands into the app’s main process, exposing stored chats and data available through connected applications. OpenAI had documented a fix on September 25, according to WIRED and OpenAI’s release notes. The reporting describes a potential exploit, not confirmed data theft or exploitation in the wild.

What the report says

WIRED reported that the app used digital-signature and process checks to distinguish trusted OpenAI components from other software. Objective-See researchers found that an untrusted script could be passed through a trusted script interpreter, allowing it to reach the main ChatGPT process despite those checks. The vulnerability was local: an attacker would first need code running on the target Mac.

If exploited, the flaw could have allowed access to ChatGPT chat logs and potentially let an attacker use the app’s position to reach connected data or applications. WIRED reported that requests could appear to come from legitimate OpenAI software. Those are described capabilities, not evidence that any user’s data was stolen. OpenAI’s changelog confirms a macOS security fix, while WIRED provides the technical account and researcher attribution.

Why it matters for businesses

For a business, an AI desktop app can sit close to information employees use in daily work. Conversations may contain customer details, internal documents or business decisions. Connected applications may hold further sensitive material. The practical issue is not limited to the model’s answers: local software permissions and the way app components trust one another also affect the exposure.

The report does not establish that every ChatGPT user or organisation was affected, that the flaw could be exploited remotely without local code, or that other operating systems and products shared the issue. Businesses should not treat the report as proof of a breach. They should treat it as a reason to verify patch status and review which work data is made available to desktop AI tools.

What to do in practice

For organisations of any size, the immediate response is straightforward: confirm that Macs running ChatGPT have received the fix OpenAI documented, and make sure someone owns that check. Where devices are managed centrally, include the app in the normal software inventory and update process. Where devices are not centrally managed, ask employees who use the Mac app to check for updates and report completion.

It is also sensible to review the information and connections available to the app. Keep access limited to what the work requires, and avoid putting credentials or highly sensitive material into an AI conversation unless the organisation has approved that use. These are general access-control precautions; WIRED’s report does not say they would have prevented this specific flaw.

• Verify the ChatGPT macOS app is updated with the fix OpenAI documented.
• Check which business data and connected applications employees make available to desktop AI tools.
• Use the organisation’s usual security process if a device shows signs of unauthorised access; the report itself does not establish that a compromise occurred.

What to watch next

OpenAI’s release notes confirm that it fixed a macOS security issue on September 25, and WIRED published its account on October 2. The available reporting does not confirm exploitation in the wild or identify affected businesses. It also does not establish the full range of vulnerable versions or explain whether every user received the fix automatically. Further technical details or a fuller vendor advisory could clarify those points.

For business leaders, the useful distinction is between a potential weakness and a confirmed incident. The report supports checking and updating the app; it does not support assuming that chats were stolen. More broadly, when software is given access to business conversations and connected tools, its security and update process belong in the operational review alongside the task it automates.