GoKawiilSecurity researchers found a ChatGPT exploit letting attackers hijack accounts via a shared link
A research team discovered a vulnerability in ChatGPT's share and HTML Preview features that let attackers run malicious JavaScript inside a victim's authenticated chatgpt.com session simply by sending them a link. Once opened, the code could read the victim's private conversations and call ChatGPT's authenticated APIs on their behalf, effectively taking over the account without any further action from the victim. OpenAI fixed the issue on October 26 and classified it as a P1 (critical) vulnerability.
GoKawiil's interpretation of the reporting above, not reported fact.
The flaw shows how combining two seemingly benign features—public sharing and HTML Preview—can create a serious cross-account attack surface in widely used AI products. Because the exploit required no downloads, code pasting, or developer tools, it suggests that even sophisticated users could be at risk from a single malicious link, underscoring the need for rigorous sandboxing in AI chat interfaces. The researchers' disclosure and OpenAI's quick patch indicate the industry is still working through security assumptions unique to conversational AI platforms.
- A single shared ChatGPT link could let attackers execute code in a victim's authenticated session.
- The exploit allowed reading private conversations and calling ChatGPT APIs as the victim, amounting to full account takeover.
- OpenAI patched the vulnerability on October 26, 2026, rating it P1 severity.
Yubico YubiKey 5C NFC — Account takeover stories like this highlight why relying solely on passwords and session tokens is risky. A hardware security key adds a strong second factor for your critical accounts, making it much harder for attackers to hijack sessions even if other vulnerabilities are exploited. It's a simple, durable way to harden your overall account security posture.
See Yubico YubiKey 5C NFC on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: parameter.ai, 2026-10-31
Published there as: “We Took over ChatGPT Accounts”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.