Warp sends a terminal session to LLM without user consent

60 ykurtov 23 8/19/2025, 4:37:32 PM
Wonder how being used feels like? One way to experience that is to discover that your terminal silently started to send command outputs to LLMs.

Today, I got an LLM suggestion on how to fix a syntactic error after following an attempt to run a test.

So, I went on to Warp's Discord to ask what's going on, and sure enough, their "Friendly support bot" and I discovered that.

> Warp has introduced features like Prompt Suggestions and Next Command that use LLMs to provide contextual suggestions. These features are part of Warp's Active AI system, which proactively recommends fixes and next actions based on your terminal session, including errors, inputs, and outputs.

"Proactively" here also means without explicit user consent.

I did enjoy Warp, but that breach of trust is so enormous I'm removing it just now.

This tells volumes about ethics and what's important.

Ref: https://docs.warp.dev/agents/active-ai

Comments (23)

zenethian · 2h ago
I tried Warp once. Well, tried to install it. But when I ran it, instead of giving me a terminal, it showed me a login prompt. Perhaps I mistakenly downloaded a login prompt app instead of a terminal app? ;) Either way, I promptly uninstalled it as I needed a terminal app and not whatever that was.
drewbitt · 1h ago
They removed the login requirement. You can now use it without logging in. It was stupid.
serjester · 2h ago
It’s absolutely insane they managed to raise 50M dollars. Now they’re shooting for a desperate AI pivot that no one wants.
michaelbuckbee · 1h ago
One of the really nice things about Cloud Code is that it can interact with your file system, github api, local utilities, etc.

Warp feels like it's at a similar spot with their agent albeit with less Anthropic secret sauce.

dfltr · 2h ago
Anyone got a recommendation for a replacement? I'm currently using Warp and the history/context-aware autocomplete-on-meth is nice, but I don't use any of the new agentic features.
abnercoimbre · 2h ago
Self-plug: Still in early days, but I've positioned myself as the indie alternative. Terminal Click [0] is a 100% offline binary for Windows, Mac and Linux. I'm also blurring the lines between plain text and rich graphics. Zero AI is involved.

Consider subscribing to the RSS feed [1] at the very least :)

[0] https://terminal.click

[1] https://terminal.click/index.xml

zenethian · 2h ago
Fish shell has this capability and you don't need any special terminal.

I imagine there are zsh scripts and/or omz plugins for it too.

iTerm2 on Mac has extra integrations also.

dahjelle · 1h ago
Do you mean iTerm2 extra integrations with fish shell or that it has a setting somewhere that integrates Warp’s fancy GUI autocomplete stuff?
beacon473 · 23m ago
https://github.com/wavetermdev/waveterm

I'm a Warp fanboy. Claude Code has it beat for writing software, but Warp is magic for linux sys admin. I SSH into my home server and feel like a wizard, no more constantly switching to a web browser to Google stuff. The experience of staring at a text only terminal for hours without ever switching to a different window feels like using DOS before the internet. It's magical.

closeparen · 1h ago
Isn't the express purpose of Warp to be an AI-centric terminal? What's the use case for doing stuff in Warp that you don't want sent to an LLM vs. using a regular terminal?
nerdjon · 1h ago
I am normally pretty quick to jump on how bad for privacy shoving some random LLM tool into a product is and the serious security risks especially in the terminal...

But looking at the marketing for Warp, this thing screams LLM everything. Nothing about this hints that things are processed locally. I can't imagine using a tool like this and not thinking that everything I type into it (and give it access too) is getting routed to a server somewhere.

What am I missing here about being upset that... it seems to be doing its job?

Unless I am missing that it is installing something so this happens in your normal terminal or something like that... to be blunt if you used this tool and this is what breaks your trust... how did you think it worked in the first place?

drewbitt · 1h ago
Yes, and even going back over 2 years ago in Wayback Machine has Warp advertising AI autocorrect as one of its biggest features.
thehamkercat · 56m ago
Terminal is the only app where you don't want any AI/Cloud integration, trust me
egamirorrim · 1h ago
Yeah I had such high hopes for Warp before it launched and then it's slowly enshittified. Turns out Ghostty was what I wanted all along.
xyst · 3m ago
Warp must be hemorrhaging all of their cash if they stoop so low as to try to scam _developers_

Bye bye, Warp. Trash product. Trash leadership.

drewbitt · 3h ago
Both Prompt Suggestions and Next Command are settings you can disable.
hu3 · 2h ago
should be disabled by default.

Or present a big warning before enabling.

I often set secret tokens as env variables, even if temporarily when running commands.

There's no way I'm touching warp with a ten foot pole after that.

bigyabai · 4h ago
Warp has been a disaster product for years. Click into almost any of their HN threads and you'll find damning condemnations of their sneaky behavior: https://hn.algolia.com/?dateRange=all&page=0&prefix=false&qu...
RadiozRadioz · 1h ago
Yes, and also condemnations about their project in general. I have noticed they have stopped commenting as much on posts that mentioned them.

Something makes me think HN might not be the target demographic for a bloaty proprietary terminal with a login prompt & LLMs stuffed into it.

rstat1 · 2h ago
I uninstalled it after finding basic terminal functions broken.

I tend to append "clear &&" to commands I run frequently, to clear out output from a previous run. Every other terminal this works like you expect. In Warp, it doesn't. Turns they've hijacked the "clear" command for reasons I don't remember, such that it only works when you run it separately instead of as part a sequence. I only learned this when I went searching for a bug report on that found one that had been opened for a while where they essentially said they had no interest in making this sort of basic stuff work.

NitpickLawyer · 4h ago
I'm sorry but this should have been obvious once the "terminal" asks you to authenticate to a remote account. Yeah, thanks but no thanks.
rvz · 4h ago
very concerning.
myflash13 · 1h ago
So does every AI agent such as Claude Code. I’ll take the privacy risk for the 10x increase in productivity. To help mitigate the security risk I will be encrypting environment files and rotating secrets far more often.