A Shared ChatGPT Link Could Run Attacker Code in Your Session and Read Private Chats

Affiliate disclosure: We may earn a commission from qualifying purchases through these links at no additional cost to you. As an Amazon Associate I earn from qualifying purchases.

A malicious ChatGPT share link could, under the conditions described by researchers, run attacker-controlled code inside a victim’s active logged-in ChatGPT session and expose private conversations, according to Parameter’s findings reported by The Hacker News on 2026-10-31.

The issue matters because it was not just a broken link or a nuisance bug. In the reported chain, a shared chat could trigger code execution in the victim’s session and potentially let an attacker act as that user inside ChatGPT, including reading conversations that were never intended to be public. The publication says OpenAI rated the issue P1 and rolled out a fix in its web frontend within eight days of disclosure (The Hacker News, 2026-10-31).

What researchers say happened

According to the report, the vulnerability depended on three design flaws that worked together to escape ChatGPT’s sandbox. ChatGPT’s Preview feature was used to isolate potentially unsafe HTML content on a separate web address, but the malicious shared chat included an interactive graph that loaded a math library called MathLive.

The key problem, as described in the source, was that MathLive ran a listener in the main ChatGPT page that accepted messages about its on-screen keyboard without verifying the sender and allowed raw HTML to be included. That let the attacker push raw HTML into the trusted main page, create a new frame the browser treated as part of chatgpt.com, and then leverage a script hosted on YouTube because ChatGPT’s content security policy permitted scripts from there (The Hacker News, 2026-10-31).

Why this is security-relevant

If accurate, the impact goes beyond a typical cross-site scripting bug because the report says the code executed within a logged-in ChatGPT session. That means an attacker could potentially perform actions on behalf of the victim and access private chats that were not shared publicly.

The reporting source attributes the discovery to security firm Parameter and says the flaw was disclosed to OpenAI on October 18, 2026. OpenAI then fixed the issue across its web frontend by October 26, 2026, and paid the researchers a $5,000 bounty, according to the same report (The Hacker News, 2026-10-31).

What we can say, and what remains limited

The source provides a detailed description of the exploit chain, but it is still a single reported account of the issue rather than independent verification here. So the safest conclusion is that, based on the published findings, a share-link-based attack path existed, OpenAI treated it as highly serious, and remediation followed quickly.

For users, the practical takeaway is simple: shared AI content can carry more risk than it appears, especially when embedded components, browser trust boundaries, and third-party libraries are involved.

Conclusion

This reported ChatGPT vulnerability shows how a seemingly ordinary shared link can become a session-level security problem when multiple web-app assumptions fail together. Parameter’s findings, as reported by The Hacker News, suggest the flaw could expose private chats inside a logged-in session, and OpenAI’s P1 response indicates the company considered it severe enough to patch quickly (The Hacker News, 2026-10-31).

Sources and dates

Research completed: 2026-11-01T05:17:03Z

Building your own website?

This separate resource is for readers creating an online business: Explore Systeme.io. It is an affiliate link, not an Amazon product recommendation.