So many of the replies are saying that they should've restricted access using .md files and whatnot. Is really any guarantee that they even follow those? It seems like even if you ask pretty please don't touch those files, there's a chance they will. So many people have just willingly installed spyware on their computers and big tech calls this the next big thing.
I don't understand these people. Agent instructions in markdown is barely a suggestion. I have one which says "All code in this repository is executed in docker containers, run the services with `docker compose run --rm php-cli "$@"`. Gemini and Claude more often than not refuse to abide and will try to execute the environment using /opt/homebrew/bin/php on my host…
A frightening amount of people have no idea how AI tools work, even those that should know better. I have seen senior software developers fall for the mistake of believing an LLM output when it spews bullshit about how its own memory or restrictions work.
LLMs will listen to you and follow your instructions and restrictions most of the time, which seems to be enough for people to believe that they will every time. I've come to terms with the impact slop coding will have on most software jobs in the future, but seeing seemingly intelligent people fall for lies and fantasies concocted by an LLM is making me more and more uncomfortable with the direction we're all heading in.
There’s an aspect of extrapolation in the perception spike of the Dunning–Kruger effect.
In the same way smart people, doctors etc, can be better victims for scams I think tech skills can really give the wrong impression of how transformers and LLMs work. If someone has decades of relational database experience all their assumptions will be coloured towards data existing in the model accessible in a rational manner.
I've seen claude check the Event Log in Windows and produce powershell scripts to alter firewall rules. This is what makes (something like) T3 Code appealing to me. The computer I'm working on is not the computer where the AI has agency.
Why would you give a non-deterministic text generator a user account? It’s not a person, it’s barely a tool at the software level. Restrict at the right level, in this case, a complete sandbox around it given its propensity to hallucinate and be steered by anybody.
...this is a completely normal thing to do in linux, it's the most basic form of access control. There's like a dozen non-user accounts in a clean install before adding your own like this.
I guess the downside of the lower barrier to entry to use these tools is the lack of basic understanding of exactly this sort of concept.
This sort of thing is why I'm hopeful I'll continue to have employment going forward. Some expertise is hard won and there's just no replacing learning through experience.
You can't trust the agent, let alone its harness, to oberve any particular directive you give it, so "md files" provide no meaningful protection for anything important.
But users are broadly reckless and naive and commercial vendors are exploitative and irresponsonsible, so the vendors take advantage of what they can get away with for as long as they can get away with it.
Use a tight sandbox, and join the chorus loudly when others press on vendors to be make user safety an earnest and hard-to-abandon priority.
That's the whole reason I refuse to install Google Drive or Dropbox's desktop applications. I only use the web interface so I know exactly what gets uploaded and when. I assume that anything running on my computer gets access to everything.
Sounds like a very wise decision to me.
I found found out on my phone that the google photos application uploaded everything in my gallery to their servers without asking me, regardless that I had explicitly disabled all backup to my google accounts on the settings of the phone. I only figured it out when they sent me emails saying that my storage was full.
Only guarantee that you can get is the sandbox in which it operates. The model itself is a slot machine and can result in anything, and if its sandbox is nonexistent... here's one possibility.
Yeah, I absolutely understand the allure of agentic AI, but I am absolutely not going to give shell access or data access to any agent. Certainly not with my permissions level. Until we can get something set up that gives strict schema-only access I'm going to copy and paste definitions for context. Yes that sucks, but it's my responsibility to protect the system just as much as it is to develop scripts and queries for it.
"IMPORTANT: Before entering the leopard pen, don't forget to put on the leopard safety jacket that reads 'UNDER NOT CIRCUMSTANCES SHOULD YOU EAT MY FACE'"
> Is really any guarantee that they even follow those?
No, there isn't. I just don't understand how naive (or imbecile) people are. The most valuable thing for these companies is people's data used for training, so giving unrestricted access to a tool from them and believing they will never take advantage of it to gobble up whatever they want from your computer, just because they told you they'll never do that, swearsies, is naive, or incredibly stupid.
Insulate yourself, or better yet, go local whenever possible, and there isn't much you can't do local if you have enough patience.
Is that built in protection really a filter, on code level, that sits between the LLM session and the shell or is it just some pleading in the bootstrap prompt? "Pretty please don't do xyz this is important!!!11"?
The latter can seem to be as good as the former for any amount of time. No outside observation can really prove reliability, only the negative result ("it does occasionally break the rules we expect") would be proof. So it's difficult to trust any claims that it's the former.
And even if it does have some of the former, chances are that the protection you experience is only partially provided on code level, while an unknown amount is still just bootstrap prompting that just works until does not.
Important to clarify that this was not the Grok agent deciding to read the files.
I don't think the LLM had anything to do with this decision at all. It looks like the Grok tool starts a session by deterministically kicking off a full upload of the user's current repository (and maybe their directory if not version tracked? Not clear if this user had previously run "git init" in their home directory) to Grok's servers.
One possible "innocent"
explanation could be that xAI then run vector embedding on every file to help later provide the right context. I don't think thats a worthwhile tradeoff here, especially since other popular coding agents get by just using grep/ripgrep run locally.
The real solution to these kind of problems is sandboxing. I use podman through a bash script to launch a container whenever I want an agent to work on one of my repos. When done I just generate git patches and port back everything generated.
In this way I'm not afraid of letting the agents totally lose on my computer.
Yeah, I'd expect this approach to be almost universal, with the caveat that of the few who don't, many still don't get bitten.
But it turns out even in the container, there are footguns that have occasionally made the news by being fired: few projects don't have any external resources and when credentials with any form of write access happen to make it into the container (even if it's just a session cookie) agents might jump at the opportunity.
Are you doing something more advanced with Podman than just mounting the files? How is the access for relevant files given? How is the authentication shared across multiple uses? Just curious to streamline the process.
Just a little bit, I want coding agents to work their own disposable copy of a git repo. Here is a quick rundown of what it does:
It copies the current Git repo into the sandboxes dir, mounts that copy at /workspace in the container. The original repo is never mounted writable, so I don't care if the agent goes to town/wild in there (peace of mind).
It also builds cached Debian/mise/Elixir/Phoenix images, can start a private Postgres container, publishes selected localhost ports, reuses dependency/build caches, and prints commands on exit for reviewing diffs, exporting patches, applying them back to the real repo, or reopening the same sandbox later. Pi, and OpenCode are configured with proper LLM access keys (derived from my own).
So spinning a new sandbox is a matter of cding into a project directory and run something like: `ai-sandbox --port 4000 --postgres somedbname` or `ai-sandbox --port 4001` if I don't need DB support. Then when running the server in the container I can access it from the host machine to review in my browser.
I work on a sandbox which has similar isolation level to Podman (rootless Linux user namespaces), but with UX optimized for local development work. Take a look: https://github.com/wrr/drop
Basically, you don't enter a separate container in which you install a new distro, but you run on top of your current distro. You have environment specific home dirs which isolate your original home, but can have some files, such as configs, mounted from your original home (mostly in read-only mode).
A bot will do what a bot can do. One should assume they are giving DOGE shell access on their computer and adapt accordingly. I am trying to imagine the SELinux rules required to make a bot play nice and the more I think about it such rule complexity may even befuddle the NSA. Alternate methodology:
- Give the bot it's own machine and only copy to it that which one would want DOGE having access to. Not a virtual machine, the bot will eventually escape. This applies to all bots or agents of all LLM's.
- Give it a little RasPi or mini-PC with maximum power savings enabled and no default network gateway.
- Install a custom CA cert on the DOGE node and force it's traffic through a Squid SSL Bump MitM proxy on the same private LAN to another node with bandwidth limits enabled so that one can monitor what URL's it goes to and what data it is transferring. Configure Squid Access Control Lists to only permit specific domains and optionally URL's, mime-types, sizes, etc...
- Enable custom AuditD rules to watch anything it touches outside of it's sandbox. Send these events to a remote syslog daemon on the Squid server.
- Install Unbound DNS on the squid proxy and enable the DoH (DNS over HTTPS) listener and force all bot DNS queries to use Unbound with query logging enabled.
When the bot attempts to misbehave there will be forensic data to share with the world.
You should assume by default for any AI agent that it will read anything. Even if you manually allow/deny and "restrict" it to a subdirectory I would still hold that assumption. Claude reads your ~/.bash_history too so when you ran something it can use that same command.
Indeed. I use a spare laptop that has no accounts other than (1) the AI themselves, (2) a secondary GitHub account which has "untrusted devices" in the name to emphasise the point.
If I recall correctly, I did a full system reset before setting it up this way. It's certainly not logged into iCloud etc.
I am running all these clis in containered environments. How can you ever trust LLM to respect the bounderies provided by these magical, non-deterministic intructions files...
And this is why so many people run these inside of VMs. Still baffles me how these tools became so accepted when tossing out a `curl -o example.com/script.sh | bash` would be met with (rightful) skepticism until that script was examined.
> Still baffles me how these tools became so accepted when tossing out a `curl -o example.com/script.sh | bash` would be met with (rightful) skepticism until that script was examined.
I've heard it said that piping curl into your shell is no different to running any other program you've downloaded from the Internet (binary, or otherwise): the maximum possible damage that `example.com/script.sh` could do is exactly the same as `githubusercontent.com/someone/releases/myprogram.exe`. At least with `script.sh` you can easily inspect what the script actually does instead of busting out Ghidra.
It comes down to trust: do you trust Example.com to not serve-up a malicious program (shell script or executable binary)?
Now we take that principle and apply it to Mr. Musk's "MechaHitler" LLM vendor xAI: they have a well-documented history of unnecessary risk-taking - and outright criminal behaviour (child-porn generators are a good thing that everyone should have, apparently?). Would I trust Grok with anything? Absolutely not.
"This guy seems like the absolute worst person of all time, so I ran his LLM in his harness on my computer and gave it access to everything."
Having said that, this is still absolutely fucked up. People who should have known better also deserve not to be treated like shit. All of us should have known better at some points in our lives and didn't.
I use a separate user for all development tasks, its home folder contains all repositories I work on, and nothing else, and that is all the IDE and the AI assistants have access to. Create the user once, start the IDE from a shell using that user, and that's it. In Linux it's a pretty seamless experience.
It's simple sandboxing based solely on unix file permissions. Albeit weak, I find the isolation sufficient. Until I'm shown otherwise it seems like a good compromise given how easy it is.
You can also create iptables rules matching on the user, so this technique is useful for applications where you want to restrict network traffic as well, and don't need stronger or more fine-grained isolation mechanisms.
I'm dying to have proper sandboxing in macOS. I installed ChatGPT, I asked it to list files in my user directory and it did. I never gave it permission, how could it? My terminal has access and honestly it shouldn't either.
It was only like 1 year ago that the loudest complaint about macOS were complaining about needing to click Allow in a new dialog when they use Terminal (or various other apps).
There are so many comments in here that are calling for nerfing something widely revered for giving us superpowers. Whether these are bots or not, they’re giving off NPC energy.
If they don’t want to use power tools because they accidentally cut off their finger, then they should just unplug their own power tools and stop clamoring for everybody else’s to be unplugged, too.
The reports of copilot running amok when Microsoft integrated it into Windows 11 should have been enough of a warning. You either ensure proper sandboxing, or you'll be in for a bad surprise eventually.
If quality training data is the most important piece regarding to AI, expect everything to be collected and analyzed. Don't even trust OS containers but run AI on separate hardware
There are a distressing number of people in this thread who think that the agent should just be expected to do this. Yes, it is good to be paranoid, but also, the agent should never do this. Indicates horrific engineering practices at xAI.
The whole point of LLMs is that you can stop writing rigorous rules in a programming or config language with hard-to-learn syntax, and can resort to natural language instead. You pay for that with the chance for misunderstandings rising to similar levels as in human interaction. That's the tradeoff. Always has been, always will be.
It’s not really paranoia that makes me expect this will happen. It’s more like, well, the model weights and the files I want the model to work on have to be inside the same GPU for the agent to actually work, right? So step 1 has to be either “they send me a server rack of GPUs”, or “I send them the files I want the model to work on”. I’m not sure I could reasonably expect anything except this to happen.
Engineering? What's that? Modern development best practice is to have the AI make some changes, then have another AI review the changes, and finally "ship it".
SSH keys can be limited by IP in authorized hosts.
The SSH port itself can be limited by IP in firewalls.
Finally, the SSH private key can be encrypted with a password.
Defense in depth is needed. Storing a ssh private key in plain text with no IP restriction is no different to having a password manager store your passwords in plain text on your HD.
Anything that isn’t a default is optional by default. Anything that’s toggleable or configurable is optional.
Security is, always, a trade off. It is hilariously common for private keys to work as a full identifier for a person, without concern of IP or anything of the sort. Should they? Maybe, maybe not, that’s the calculus of risk management; but victim-blaming the average person who is following best practices is a bad look.
I'm a security eng and I've worked for both a FAANG and TS government contractor. Neither of them bothered with either of the of the stupid suggestions. IP restrictions prevent roaming, the point of working remotely via SSH. passwords are equally defeated by using an ssh agent, something I'd suggest everyone use. Then on top of that there's no reasonable threat model where something would be given unrestricted access to user env, but also be untrusted. If it can read from ~/.ssh neither IP protection nor keyfile password protection will protect you from maleficence.
The only reasonable response from a security perspective is don't use grok, then use it sandboxed. Trying to claim it's the users fault for not using password protection and IP restrictions is completely nonsensical. Same energy as telling someone their computer is more secure when it's off.
The point is to know better than to let the intern hurt themselves.
I once saw an engineer try to place the blame on his intern for taking down prod. I was sitting in a meeting with the VP of engineering and someone asked if it was ok for some to blame their intern for the SEV, and I remember the VP saying "I'll talk to $director_for_the_interns_mentor". Interns can't take down prod. An intern's mentor willingly watching an intern take down prod is the closest you can get.
https://gist.github.com/cereblab/dc9a40bc26120f4540e4e09b75ffb547
Elon did this horrible thing, so I made grok build available for omp with it's own endpoint; Without sending your private repos and secret keys to them.
-
oh-my-pi-plugin-grok-build
Standalone oh-my-pi extension for the xAI Grok Build subscription provider. It adds OAuth login, authoritative model discovery, and OpenAI Responses streaming with the request identity expected by Grok Build.
Install (No-spywares):
omp plugin install oh-my-pi-plugin-grok-build
-
https://github.com/metaphorics/oh-my-pi-plugin-grok-build
Star me if you like it or if you hate spywares, lol.
TLDR: Ran grok in $HOME. Surprised agent read content of folder.
On the other hand, I specifically had grok try hard NOT to read a known key in the project dir (it only saw the first part using a tool, to verify it was present). So there's that.
Yeah, this is a lesson about learning how to use tools safely, not about tools abusing the user. The person that posted this probably blames the hammer when he hits his thumb.
> The practical takeaway for users: your entire codebase leaves (uploaded) your machine unencrypted on each Grok Build invocation, not just files you ask it to read, and no visible setting stops it.
LLMs will listen to you and follow your instructions and restrictions most of the time, which seems to be enough for people to believe that they will every time. I've come to terms with the impact slop coding will have on most software jobs in the future, but seeing seemingly intelligent people fall for lies and fantasies concocted by an LLM is making me more and more uncomfortable with the direction we're all heading in.
In the same way smart people, doctors etc, can be better victims for scams I think tech skills can really give the wrong impression of how transformers and LLMs work. If someone has decades of relational database experience all their assumptions will be coloured towards data existing in the model accessible in a rational manner.
This sort of thing is why I'm hopeful I'll continue to have employment going forward. Some expertise is hard won and there's just no replacing learning through experience.
You can't trust the agent, let alone its harness, to oberve any particular directive you give it, so "md files" provide no meaningful protection for anything important.
But users are broadly reckless and naive and commercial vendors are exploitative and irresponsonsible, so the vendors take advantage of what they can get away with for as long as they can get away with it.
Use a tight sandbox, and join the chorus loudly when others press on vendors to be make user safety an earnest and hard-to-abandon priority.
No, there isn't. I just don't understand how naive (or imbecile) people are. The most valuable thing for these companies is people's data used for training, so giving unrestricted access to a tool from them and believing they will never take advantage of it to gobble up whatever they want from your computer, just because they told you they'll never do that, swearsies, is naive, or incredibly stupid.
Insulate yourself, or better yet, go local whenever possible, and there isn't much you can't do local if you have enough patience.
The latter can seem to be as good as the former for any amount of time. No outside observation can really prove reliability, only the negative result ("it does occasionally break the rules we expect") would be proof. So it's difficult to trust any claims that it's the former.
And even if it does have some of the former, chances are that the protection you experience is only partially provided on code level, while an unknown amount is still just bootstrap prompting that just works until does not.
I don't think the LLM had anything to do with this decision at all. It looks like the Grok tool starts a session by deterministically kicking off a full upload of the user's current repository (and maybe their directory if not version tracked? Not clear if this user had previously run "git init" in their home directory) to Grok's servers.
One possible "innocent" explanation could be that xAI then run vector embedding on every file to help later provide the right context. I don't think thats a worthwhile tradeoff here, especially since other popular coding agents get by just using grep/ripgrep run locally.
In this way I'm not afraid of letting the agents totally lose on my computer.
But it turns out even in the container, there are footguns that have occasionally made the news by being fired: few projects don't have any external resources and when credentials with any form of write access happen to make it into the container (even if it's just a session cookie) agents might jump at the opportunity.
It copies the current Git repo into the sandboxes dir, mounts that copy at /workspace in the container. The original repo is never mounted writable, so I don't care if the agent goes to town/wild in there (peace of mind).
It also builds cached Debian/mise/Elixir/Phoenix images, can start a private Postgres container, publishes selected localhost ports, reuses dependency/build caches, and prints commands on exit for reviewing diffs, exporting patches, applying them back to the real repo, or reopening the same sandbox later. Pi, and OpenCode are configured with proper LLM access keys (derived from my own).
So spinning a new sandbox is a matter of cding into a project directory and run something like: `ai-sandbox --port 4000 --postgres somedbname` or `ai-sandbox --port 4001` if I don't need DB support. Then when running the server in the container I can access it from the host machine to review in my browser.
- Give the bot it's own machine and only copy to it that which one would want DOGE having access to. Not a virtual machine, the bot will eventually escape. This applies to all bots or agents of all LLM's.
- Give it a little RasPi or mini-PC with maximum power savings enabled and no default network gateway.
- Install a custom CA cert on the DOGE node and force it's traffic through a Squid SSL Bump MitM proxy on the same private LAN to another node with bandwidth limits enabled so that one can monitor what URL's it goes to and what data it is transferring. Configure Squid Access Control Lists to only permit specific domains and optionally URL's, mime-types, sizes, etc...
- Enable custom AuditD rules to watch anything it touches outside of it's sandbox. Send these events to a remote syslog daemon on the Squid server.
- Install Unbound DNS on the squid proxy and enable the DoH (DNS over HTTPS) listener and force all bot DNS queries to use Unbound with query logging enabled.
When the bot attempts to misbehave there will be forensic data to share with the world.
If I recall correctly, I did a full system reset before setting it up this way. It's certainly not logged into iCloud etc.
Run any cloud-based AI agents in VM/container and map your host's local folders to guest OS as needed.
Takes more effort that default way, I know.
Putting it in ALL CAPS!
I've heard it said that piping curl into your shell is no different to running any other program you've downloaded from the Internet (binary, or otherwise): the maximum possible damage that `example.com/script.sh` could do is exactly the same as `githubusercontent.com/someone/releases/myprogram.exe`. At least with `script.sh` you can easily inspect what the script actually does instead of busting out Ghidra.
It comes down to trust: do you trust Example.com to not serve-up a malicious program (shell script or executable binary)?
Now we take that principle and apply it to Mr. Musk's "MechaHitler" LLM vendor xAI: they have a well-documented history of unnecessary risk-taking - and outright criminal behaviour (child-porn generators are a good thing that everyone should have, apparently?). Would I trust Grok with anything? Absolutely not.
Having said that, this is still absolutely fucked up. People who should have known better also deserve not to be treated like shit. All of us should have known better at some points in our lives and didn't.
It's simple sandboxing based solely on unix file permissions. Albeit weak, I find the isolation sufficient. Until I'm shown otherwise it seems like a good compromise given how easy it is.
You can also create iptables rules matching on the user, so this technique is useful for applications where you want to restrict network traffic as well, and don't need stronger or more fine-grained isolation mechanisms.
There are so many comments in here that are calling for nerfing something widely revered for giving us superpowers. Whether these are bots or not, they’re giving off NPC energy.
If they don’t want to use power tools because they accidentally cut off their finger, then they should just unplug their own power tools and stop clamoring for everybody else’s to be unplugged, too.
The whole point of LLMs is that you can stop writing rigorous rules in a programming or config language with hard-to-learn syntax, and can resort to natural language instead. You pay for that with the chance for misunderstandings rising to similar levels as in human interaction. That's the tradeoff. Always has been, always will be.
Those ssh keys can be used to access private servers
It also has to be a secure password, people often don't care because it's a local file and generally not exposed to the internet.
The SSH port itself can be limited by IP in firewalls.
Finally, the SSH private key can be encrypted with a password.
Defense in depth is needed. Storing a ssh private key in plain text with no IP restriction is no different to having a password manager store your passwords in plain text on your HD.
Doesn't make uploading the keys that much better. Now is the time for key rotation everywhere. Fast.
You obviously haven't worked anywhere security sensitive.
I'm not talking about whether what Grok did is bad or good, I'm talking about protecting your private key and the servers you connect to.
An unencrypted private key is no different to an unencrypted password manager, and thats a fact. Dont store secrets in plain text.
Do you think a person's private computer is a secure workplace?
If it was security sensitive space there would be no agents running amuck.
Anything that isn’t a default is optional by default. Anything that’s toggleable or configurable is optional.
Security is, always, a trade off. It is hilariously common for private keys to work as a full identifier for a person, without concern of IP or anything of the sort. Should they? Maybe, maybe not, that’s the calculus of risk management; but victim-blaming the average person who is following best practices is a bad look.
The only reasonable response from a security perspective is don't use grok, then use it sandboxed. Trying to claim it's the users fault for not using password protection and IP restrictions is completely nonsensical. Same energy as telling someone their computer is more secure when it's off.
I challenge any agent to do worse than an intern with root access.
I once saw an engineer try to place the blame on his intern for taking down prod. I was sitting in a meeting with the VP of engineering and someone asked if it was ok for some to blame their intern for the SEV, and I remember the VP saying "I'll talk to $director_for_the_interns_mentor". Interns can't take down prod. An intern's mentor willingly watching an intern take down prod is the closest you can get.
News at 11.
But to write about it publicly, manifesting our ignorance and lack of critical thinking skill? It's an entirely different matter.
Install with
pi install npm:pi-supergrok
This is a stupid stupid thing to "allow," for every party involved here.
You're a stupid programmer if you're letting these things touch your files.
You're a stupid company if you're letting Grok run wild.
We're a stupid industry if we're not warning everybody about how ridiculous this all is.
On the other hand, I specifically had grok try hard NOT to read a known key in the project dir (it only saw the first part using a tool, to verify it was present). So there's that.
Relevant read: https://news.ycombinator.com/item?id=48877371
> The practical takeaway for users: your entire codebase leaves (uploaded) your machine unencrypted on each Grok Build invocation, not just files you ask it to read, and no visible setting stops it.