SDSignal Desk

Researchers found a ‘trivial’ ChatGPT Mac app flaw exposing chats and command execution; OpenAI patched it

· Updated · Signal Desk newsroom

Analysis of reporting by WIRED. How we work

Image: WIRED

WIRED reports Objective-See researchers found a vulnerability in OpenAI’s ChatGPT macOS app that could let local malware read chat logs, ride browser sessions and run commands disguised as OpenAI instructions. OpenAI fixed it September 25 and says it needs to move faster.

Key takeaways

  • Objective-See researchers found the flaw in ChatGPT’s macOS app.
  • It could expose chat logs, browser sessions and allow disguised command execution.
  • Patrick Wardle called it “insanely trivial,” about a dozen lines of code.
  • OpenAI noted the fix in its September 25 changelog.
  • Wardle says he has filed another report on ChatGPT’s Dots integration.

The vulnerability

WIRED reports a now-patched vulnerability in OpenAI’s ChatGPT macOS app discovered by Objective-See Foundation researchers. The bug could let an attacker with local access take over the app: read chat logs and other stored data, ride on browser sessions and coerce ChatGPT into running commands that appear to be legitimate OpenAI instructions.

Researcher Patrick Wardle called the exploit “insanely trivial,” roughly a dozen lines of code, achieved by nesting a script interpreter so that parent and grandparent process signature checks passed. OpenAI acknowledged the fix in its September 25 changelog. Spokesperson Shane Bauer said the company is evolving its security practices but “recognize[s] a need to move faster.”

AI clients as privileged targets

Much recent coverage concerns agents attacking outward. This is the reverse: AI clients are valuable local targets because they already hold extensive system trust and chat histories. Wardle’s point is that if an assistant needs deep system access to be useful, subverting it is especially damaging. Multi-layer signature checks that still accept nested script interpreters suggest helper processes need closer threat modeling.

Feature velocity widens the surface. Wardle says he has filed another report on ChatGPT’s integration with the new Dots assistant (Dots), and he recently reported a dictation-token mishandling issue in Meta’s Muse; Muse has had other security problems (Muse flaw). OpenAI’s app-like plug-ins and automations (plug-in expansion) add further integration points.

Security as part of the product

Desktop AI is worth having, which makes client security part of the product rather than an afterthought. A dozen-line proof of concept sits awkwardly beside a promise to move faster. Labs staffing macOS and Windows hardening at the pace of agent features, and wider public audits of AI desktop apps, would help. Wardle is due to speak at the Objective by the Sea conference in November, which could prompt broader scrutiny.

Users should update promptly and treat any local malware able to talk to a trusted AI process as a data-exfiltration path. Enterprises can apply endpoint controls and least-privilege policies to AI clients as high-sensitivity applications. More in our cybersecurity coverage, OpenAI coverage and tools coverage.

Who feels it

ChatGPT desktop users
Patch promptly; local malware could exfiltrate chats through the app.
AI platform security teams
Threat-model helper processes and signature checks.
Enterprises
Endpoint controls for AI clients matter as much as model alignment.

What to watch

  1. OpenAI’s response to the Dots-integration report.
  2. Public audits of other AI desktop apps.
  3. Enterprise guidance treating AI clients as high-sensitivity apps.

FAQ

Who found the flaw?

Researchers at the Objective-See Foundation, including Patrick Wardle.

What could attackers do?

Read chat logs, ride browser sessions and run commands disguised as OpenAI instructions.

Is it fixed?

Yes, OpenAI noted the fix in its September 25 changelog.

What did OpenAI say?

It recognizes a need to move faster on security, per spokesperson Shane Bauer.

Read the original

Continue at the source.

WIRED

Companies: OpenAI

More on OpenAI

All OpenAI coverage →