AI news · ChatGPT · shared links
ChatGPT share links could hijack accounts, Parameter says
A security firm chained two ChatGPT features to run its code in a victim’s signed-in session; OpenAI fixed the flaw eight days after the report.
Until 26 October 2026, opening a booby-trapped ChatGPT share link was enough to let an attacker’s code run inside the victim’s signed-in ChatGPT session, the security firm Parameter says in its write-up. Parameter’s code pulled up a private chat the victim had never shared, after first reading their list of conversations, and passed a piece of it to a Parameter server. OpenAI graded the flaw P1, the top of its severity scale, and had a fix out on 26 October, eight days after Parameter reported it through the company’s bug bounty programme (which pays outsiders for flaws), The Hacker News reports.
The route combined two everyday features in one shared chat. The first, Preview, shows web code from a chat as a live mini page, and ChatGPT keeps that page walled off on a different web address; Parameter says the wall held. The weak point came from the second feature, an interactive graph: to draw it, ChatGPT pulled MathLive, an outside maths library, into the main page, and MathLive’s on-screen keyboard code obeyed instructions from any frame without checking who had sent them. That gave the walled-off Preview a way to slip its own web code into the real ChatGPT page.
Parameter calls the result an account takeover. Every step ran between accounts Parameter owned, and the firm reports no real victims.
ChatGPT has a Free plan; the Go card says it “may include ads” on our 31 October read; the attack needed only a signed-in account, whatever the plan.
Nothing set a trapped link apart from a genuine one, whether it arrived in a post, a chat message or an email, so a team that swaps ChatGPT links had no way to spot it from the link alone; only the fix on OpenAI’s side closed the gap.
Sources
- Parameter: We found a gap at OpenAI. Checked
- The Hacker News: A Shared ChatGPT Link Could Run Attacker Code in Your Session and Read Private Chats. Checked
Alexandre S.
Alexandre S. is a creative ad strategist and copywriter with ten years' experience in digital marketing across multiple channels.
More AI news from
- Checkout appears in a quarter of AI agent visits, Clarity finds
- Cursor pulls a fake Claude Code update that planted SSH keys
- Google trials one Sponsored heading for shopping ads in AI Mode
- Google Search panels now offer Buy now through Universal Cart
- Bluefish turns AI answer gaps into ChatGPT and Google ad briefs
- Gemini stops tagging cited links with utm_source=gemini
- Deepgram buys Pipecat maker Daily and keeps the framework open
- Koah says gross sales on its AI chat ad network rose fourfold
- Resemble AI says bot tools missed most disguised AI agents