These pages are published after PatchSiren validates generated defensive summaries against stored public CVE and source evidence.
CVE-2026-55237 is a high-severity DOM-based Cross-Site Scripting (XSS) vulnerability in AutoGPT's signup page. Versions prior to 0.6.62 are affected. An attacker can craft a malicious link that, when opened by an authenticated user, performs a client-side redirect and executes arbitrary JavaScript in the context of their browser. This could lead to credential theft, internal network pivoting, and unauthor [truncated]
CVE-2025-32437 is a high-severity vulnerability in AutoGPT, a workflow automation platform. The vulnerability is caused by the `MediaDurationBlock` component, which downloads and stores videos in a temporary directory without proper cleanup, and the `StepThroughItemsBlock` component, which allows for unlimited loops. This can lead to a Denial of Service (DoS) attack, causing the system to run out of disk [truncated]
CVE-2025-32436 is a high-severity vulnerability in AutoGPT, a workflow automation platform. The vulnerability, patched in version 0.6.63, allows for a denial of service (DoS) attack due to improper handling of temporary files and lack of limits on disk space consumption. Affected users should be aware of this vulnerability and ensure they are running version 0.6.63 or later to prevent potential DoS attack [truncated]
CVE-2025-32424 is a high-severity vulnerability in AutoGPT, a workflow automation platform, that can lead to a Denial of Service (DoS) attack. The vulnerability exists in versions prior to 0.6.63 and is caused by the `ScreenshotWebPageBlock` storing captured screenshots in a temporary directory without limiting disk space consumption. When combined with the `StepThroughItemsBlock`, which allows unlimited [truncated]
CVE-2025-32422 is a high-severity vulnerability in AutoGPT, a workflow automation platform. The vulnerability exists in the `StepThroughItemsBlock` feature, which can be exploited to cause a Denial of Service (DoS) attack by slowly iterating and downloading relatively small files multiple times. This can lead to disk space exhaustion. The vulnerability has been patched in version 0.6.63. Users should be a [truncated]
CVE-2025-32392 is a high-severity vulnerability in AutoGPT, a workflow automation platform for creating and managing AI agents. The vulnerability exists in the LoopVideoBLock feature, which allows users to input a video file and process it without any resource allocation limits. A malicious attacker can exploit this by looping a video an unlimited number of times, generating an excessively large video fil [truncated]
A missing authorization check in AutoGPT's block execution API allows authenticated users to bypass credit consumption controls. The POST /api/blocks/{block_id}/execute endpoint executes blocks without validating or deducting user credits, enabling unlimited free execution of all block types. The credit enforcement mechanism exists only in the graph execution path (manager.py) and is never invoked for dir [truncated]
AutoGPT versions 0.6.34 through 0.6.51 use Python's pickle module for Redis cache serialization without integrity or authenticity verification. The backend serializes values with pickle.dumps and deserializes with pickle.loads without HMAC, signature, or strict schema validation. An attacker with ability to poison a shared Redis cache key can achieve arbitrary code execution in the backend container conte [truncated]
CVE-2026-33232 is an unauthenticated denial-of-service issue in AutoGPT Platform. The vulnerable download_agent_file endpoint creates temporary files for each request but does not delete them after serving them, allowing repeated requests to consume disk space until the backend becomes unavailable. The reported impact is server-wide service failure, including database or other component errors due to "No [truncated]
## Summary CVE-2026-30950 is a HIGH-severity (CVSS 7.1) authenticated session-hijacking vulnerability in AutoGPT, a workflow-automation platform for AI agents. Versions 0.6.36–0.6.50 are affected. An authenticated attacker who knows (or guesses) another user's session_id can reassign that session to themselves via the PATCH /sessions/{session_id}/assign-user endpoint. The flaw stems from an insecure direc [truncated]
CVE-2025-32425 is a medium-severity denial-of-service vulnerability in AutoGPT Platform, published 2026-05-13 and last modified 2026-05-26. The issue stems from unbounded container log growth in Docker deployments: execution output is captured to stdout/stderr and stored as container logs without size limits. Under high user access volume, logs can exhaust server disk resources, causing DoS. The vulnerabi [truncated]
