Repository navigation
Error-prone process in self-host deployment leads to API abuse #18442
Replies: 2 comments
|
Unexpected API charges are concerning. However, #18241 asked for help investigating usage without the version details, reproduction steps, or relevant logs needed for an actionable bug report. The issue template directs troubleshooting and questions about whether something is a bug to Discussions. Closing it was not evidence of a backdoor. LibreChat is provided as-is, without warranty, under a license that disclaims liability. We do not have the capacity to conduct individual incident-response investigations or determine how someone may have obtained access to a self-hosted deployment. Operators are responsible for securing their network exposure, registration settings, credentials, and access controls. That does not mean we dismiss security concerns. Unintended or unauthorized access caused by a vulnerability in LibreChat’s code is serious. Please submit evidence of such a vulnerability through private vulnerability reporting, including the affected version, reproduction steps, and relevant sanitized diagnostics. Do not share credentials or sensitive exploit details publicly. Your proposal for safer deployment defaults is concrete and can be evaluated on its merits. The allegation that developers intentionally abuse self-hosted instances, however, is unsupported by the evidence shared. Please keep the discussion factual and constructive. |
|
@danny-avila what exactly information you need to warn your users and to take actions related to stop malicious activity? The self hosted instances is abused right now. Above I've explained how to find and exploit self hosted servers. Your reaction demonstrates that you are not interested to stop this abusing of self-hosted instances. Your current activity is focused to decrease the visibility of my reports. I want so you know that your responses is recorded. So I ask you to stop to hiding the problem and stop closing the issues, because we have public copy of it.
You're right in terms of law, but you still responsible for this project and its reputation. I have no legal claims to you, but your reaction on security problems can negatively affect the reputation of LibreChat. Transparency is crucial for credibility. I don't ask you to work for free or anything like this, and I don't demand you to maintain my server. I reported a critical problem with LibreChat, this problem is caused by incorrect config enforced by the instruction published in official documentation of that project. I ask to make our discussion visible for any potential user of LibreChat, update the code to prevent the abuse of self-hosted instances, and to notify users about a problem. I don't ask you to urgently fix the problem. It's ok to don't rush with code changes, we can take time to think how to fix this problem in future. But we should warn users who self-hosted the LibreChat that in case they host LibreChat not for public - they should check their docker compose config because ports may be published into the internet and hackers can create account and drain their money. Also I propose to open my issue back, to ensure the highest visibility of this problem. It's ok if we will have an issue that will never be fixed, in case this issue is visible, it's a way to admit the unsolved problem and to inform potential users about it. |
Uh oh!
There was an error while loading. Please reload this page.
The problem
The current way to deploy self-hosted instance of LibreChat is error-prone, since it propose to override compose config with a file
docker-compose.override.ymland it is trivial to leave ports published for the internet.The docs page proposes to use an override for ports section:
The reputation
Recently I've filed issue about money draining through LibreChat: #18241
That issue have been closed with no answer that is suspicious.
Such fragile design and default config that allows to create a new accounts and use connected AI services through LibreChat makes me think this is intentional backdoor to let LibreChat developers abuse self-hosted instances as a free proxies for expensive LLMs.
Critical severity
I consider this problem as a security issue, since right now bots can scan IP ranges for default port
3080, then create accounts automatically and use all found servers as a free distributed proxy for LLMS.The proposal
I propose to improve deployment process to prevent such problems automatically.
LibreChat must not to publish any ports by default.
Also I propose to notify about this problem all users of LibreChat with self-hosted instance and give them instruction how to unpublish ports and how to check if their instance have been abused.
Also we must remind to users that in case their instance have been abused, the attackers can potentially upload illegal materials on the servers.
All reactions