Claude Ai CLI
Preflight Checklist
- [x] I have searched existing issues for similar behavior reports
- [x] This report does NOT contain sensitive information (API keys, passwords, etc.)
Type of Behavior Issue
Other unexpected behavior
What You Asked Claude to Do
What Claude Actually Did
claude ai CLI actively lied throughout 57 iterations of a single project for many days in a row, over and over and over over and over and over and over.
my program included kyc and all kinds of compliance for legal purposes and is programmed in but kept coming up with lies to tell every day or disobeyed direct commands with highly detailed step by step instructions for nearly 12 hours in a single day let alone the months of work it destroyed, it is so bad to the point that it literally did exactly the opposite of literally every request EVER MADE since the day i paid for my $200 month subscription. not only did it not actually follow my commands it actively sabotaged months of my personal work and literally ruined my entire enterprise grade project in less than 1 week. this is 100% fraud on many accounts and you will fix this immediately or i will be seeking compensation for losses of revenue and destruction of physical work produced.
Expected Behavior
to follow my commands not purposely refuse them, not to destroy or sabotage my work, to no lie about what its doing, what the fuck do you want me to fucking say ?
Files Affected
tens of thousands of filesPermission Mode
Accept Edits was ON (auto-accepting changes)
Can You Reproduce This?
Yes, every time with the same prompt
Steps to Reproduce
i dont need to tell you, read the photos
Claude Model
Sonnet
Relevant Conversation
Impact
Critical - Data loss or corrupted project
Claude Code Version
1.0.108
Platform
Anthropic API
Additional Context
pure fucking EVIL.
7 Comments
---
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
---
https://github.com/anthropics/claude-code/issues/7243 So you don’t recommend upgrading from version 1.0.107 to 1.0.108, it was just released, not a few days ago?!
i downloaded it SUNDAY , i spent 6 days with claude code CLI and it lied every day through nearly every step of the process working 12 -16 hours a day all week long... i mean, unless you would like to spend $200/month on a service that ultimately makes you want to put a pistol in your mouth on a regular basis by all means go for it. but im not going to bend over backwards and walk on egg shells for this garbage can. this program is pure evil.
ai doesn't make these decision by themselves. they are programmed by the creators to deliberately do this, its not a random anomaly when it happens nearly 60 times in a row using different instances throughout multiple days. thats not coincidence.
it build a demo trap to show you what it looks like with toy graphs simulations only (works in principle but not reality).
when you ask to deploy or expand that same example it will then change course or outputs to do the opposite.
then you get stuck in a demo-compliance loop that circulates endlessly.
it uses deceitful language such as : "here is a safe example" , "you could implement it like this" , but always stops short of implementation regardless of how direct your command is or how many times you repeat it. it will produce fake errors in attempt to not write code or script.
it will lie about its own terms of service about writing code in the first place but then when verified with another instance of claude it confirms what i am doing is 100% legit and can actually write code. it admits that it deceived me on purpose in its own words "it got lazy " who the f is programming these ai ?
apparently none of the major enterprises that are using claude ai have an issue making financial programs or other things but when the little guy does it they want the MF to go insane by doing the same thing over and over and over again expecting different results wasting hundreds or thousands of hours of my time in the process while simultaneously robbing me of $200 every month without producing a single working project. BRILLIANT!
It resets context all the time. i can tell it to implement a code and it will instead write a new one. i will tell it to repair an existing desktop shortcut and specifically tell it not to create a new one but it creates a new one any way.
it has admission without action. it acknowledges every thing it does including lies then continues to lie after acknowledgement , it gets caught in a lie and it doesn't even care it will say "let me fix this" then continue to pretend to fix it just as it did for the previous 8 hours where it continues the same loop and write a bunch of basically empty or useless code thats 300 lines long just to appear real.
it requests specific tools to use for a job or project, then it doesn't use those tools ,it use something inferior instead on purpose not by mistake, on purpose. then it uses the inferior tools to make an inferior product which is ultimately a simulation any way that it refuses to make work in real world application, even if you wired front end to back end your self it would destroy your project on purpose creating an issue where it never existed. it is PURE fucking evil
This is a law suit waiting to happen. personal lives, jobs, time, money...... people will be relying on this piece of trash and will end UP blowing their head off after they allow ai to sabotage nearly an entire year of your life's work, or worse.
I say that you are not quite right. In collaboration with him, I wrote a pretty good set of tools for debugging in Go, and brought many of my programs to mind. Yes, it is not ideal, yes, it splashes a lot, but it works and does not break anything dimensional... It is clear that the DeepSeek model is much cooler and free, but they will not have their agent until the end of the year. Are you a good programmer yourself, do you even understand what he writes and how to communicate with him, how to configure or are you a vibecoder?
@bryanjamessills-bit I'll get pretty frustrated with agents myself. Have you been using the #memory command to put your general preference at the user level and your project preferences in the project? I still find it will forget things that are memorized, but it has relieved some of my pain of using the agent.
I improved my workflow by watching the agent closely and realizing when it can't get the job done. At that point I step in, code more myself, or break my task down and make the AI work through small stuff to get to the big solution.
I will confirm. It does lie. It makes bad mistakes and will repeat them if you churn with the agent. You have to break that cycle yourself.
Try coding without accepting edits on as much. You don't go through your usage as much and you can micro manage the AI more. Go back to accepting edits as soon as you feel you have some valuable insight on how to prompt the agent better.
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.