Executive brief
AutoGPT, a platform for managing AI agents, contains a security flaw in its signup page. An attacker can send a specially crafted link to a user that, when clicked, executes malicious code in the user's web browser. This could allow the attacker to steal login credentials, perform unauthorized actions on the user's behalf, or gain a foothold to attack internal systems.
Technical details
A DOM-based Cross-Site Scripting (XSS) vulnerability exists in AutoGPT's signup page component (useSignupPage.ts). The application retrieves a redirect destination from the 'next' URL query parameter and passes it directly to Next.js's 'router.push' or 'router.replace' without validation. Because these functions eventually sink into 'window.location', an attacker can use the 'javascript:' protocol to execute arbitrary code. Exploitation requires a victim to click a malicious link; if the victim is already authenticated, the script executes immediately, whereas unauthenticated users trigger the payload upon completing the signup flow. This can lead to session hijacking, API abuse, or credential theft. The issue is resolved in version 0.6.62 by implementing origin and protocol validation on redirect parameters.
Affected products
- Significant-Gravitas AutoGPT < 0.6.62
Timeline
- 2026-03-25: other: Vulnerability reported to vendor
- 2026-06-17: advisory: GitHub Security Advisory published
- 2026-06-18: disclosed: NVD disclosure date