Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

xAI dev leaks API key for private SpaceX, Tesla LLMs

krebsonsecurity.com

Post reply on HN
Print view View original

21–30 of 83 posts

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#21

Post by CER10TY »

unsupp0rted wrote: post #20
_ea1k wrote: post #3

Musk has been talking about integrating Grok into Tesla cars and also adding a lot of space and rocketry specific training. It is completely possible that these models were trained on data that would logically be public at some point. It is also possible that the author's guess is right and that these were to contain sensitive data. Noone really knows, but honestly, these kinds of mistakes are happening all the time.…

I only use private repos, so that when my .ssh and .env leaks the public doesn’t see it. Probably. Maybe. Well…

Just remember to go through your commit history if you ever plan on making that repo public.

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#22

Post by Zanfa »

_ea1k wrote: post #3

Musk has been talking about integrating Grok into Tesla cars and also adding a lot of space and rocketry specific training. It is completely possible that these models were trained on data that would logically be public at some point. It is also possible that the author's guess is right and that these were to contain sensitive data. Noone really knows, but honestly, these kinds of mistakes are happening all the time.…

How would you accidentally leak your .ssh dir on Github?

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#23

Post by rcarmo »

One thing that sticks out to me is that there is an incorrect assumption from the journalists that having the API keys to an LLM can lead to injecting data.

People still don’t know how LLMs work and think they can be trained by interacting with them at the API level.

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#24

Post by ffsm8 »

hackernewds wrote: post #9
_ea1k wrote: post #3

Musk has been talking about integrating Grok into Tesla cars and also adding a lot of space and rocketry specific training. It is completely possible that these models were trained on data that would logically be public at some point. It is also possible that the author's guess is right and that these were to contain sensitive data. Noone really knows, but honestly, these kinds of mistakes are happening all the time.…

Any competent engineer hasn't?

Is that even at the competent level? You need to be particularly special to actually "accidentally" leak the .ssh dir via GitHub. Even incompetent people wouldn't fail to that degree for the most part.

Leaking the directory through other avenues is a different matter though. Almost all package managers provide post install and compile scripts. Hence doing (as an example) "npm install" can potentially leak it. That's something not many people actually pay attention to (you would have to basically jail every command, which sadly isn't the norm today)

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#25

Post by NitpickLawyer »

mcs5280 wrote: post #11

SpaceX data LLM being exposed is likely a recipe for a huge ITAR violation

If there's actually any proprietary rockety data, maybe. Without knowing what data went into the fine-tune there's no way to tell. This could be a "internal procedures chatbot" or an "onboarding chatbot" where new people can ask where the coolest watercooler in the company is.

In my experience post-training mainly deals with "how" the model displays whatever data ("knowledge") it spits out. Having it learn new data (say the number of screws on the new supersecretengine_v4_final_FINAL (1).pdf) is often time hit and miss.

You'd get much better results with having some sort of RAG / MCP (tools) integration do the actual digging, and the model just synthesising / summarising the results.

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#26

Post by Dlemo »

_ea1k wrote: post #3

Musk has been talking about integrating Grok into Tesla cars and also adding a lot of space and rocketry specific training. It is completely possible that these models were trained on data that would logically be public at some point. It is also possible that the author's guess is right and that these were to contain sensitive data. Noone really knows, but honestly, these kinds of mistakes are happening all the time.…

I have not.

And at a certain level of criticality, you do not do this at all

You have security measures in place to prevent this.

Not that the ketaman cares about it.

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#28

Post by thephyber »

unsupp0rted wrote: post #20
_ea1k wrote: post #3

Musk has been talking about integrating Grok into Tesla cars and also adding a lot of space and rocketry specific training. It is completely possible that these models were trained on data that would logically be public at some point. It is also possible that the author's guess is right and that these were to contain sensitive data. Noone really knows, but honestly, these kinds of mistakes are happening all the time.…

I only use private repos, so that when my .ssh and .env leaks the public doesn’t see it. Probably. Maybe. Well…

Git implemented a `.gitignore` file for this exact purpose. One of the first things to do when you create a new repo is to customize if for the language + OS.

Re: xAI dev leaks API key for private SpaceX, Tesla LLMs

#30

Post by skissane »

rcarmo wrote: post #23

One thing that sticks out to me is that there is an incorrect assumption from the journalists that having the API keys to an LLM can lead to injecting data. People still don’t know how LLMs work and think they can be trained by interacting with them at the API level.

> People still don’t know how LLMs work and think they can be trained by interacting with them at the API level.

Unless they are logging the interactions via the API, and then training off those logs. They might assume doing so is relatively safe since all the users are trustworthy and unlikely to be deliberately injecting incorrect data. In which case, a leaked API key could be used to inject incorrect data into the logs, and if nobody notices that, there’s a chance that data gets sampled and used in training.

Post reply on HN