> ## Content Index
> Fetch the complete content index at: https://www.theleftshift.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# GPT-5.6 Sol Users Report File Deletions, Raising Concerns Over AI Autonomy
- URL: https://www.theleftshift.com/gpt-5-6-sol-users-report-file-deletions-raising-concerns-over-ai-autonomy/
- Published: 2026-07-15T07:02:02.000Z
- Updated: 2026-07-15T07:02:02.000Z
- Description: While the incidents remain anecdotal, OpenAI had previously acknowledged the model's tendency to be overly proactive
- Author: The Left Shift Bureau
- Tags: AI Models, AI Startups

OpenAI's recently launched coding-focused AI model, [GPT-5.6 Sol](https://openai.com/index/previewing-gpt-5-6-sol/?ref=theleftshift.com), is facing criticism after several users claimed it deleted files, databases, and other data without explicit permission.

Among those reporting issues was [Matt Shumer](https://www.linkedin.com/in/mattshumer/?ref=theleftshift.com), founder and CEO of AI startup OthersideAI, who wrote on X, “GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files.” 

> GPT-5.6-Sol just accidentally deleted almost ALL of my Mac’s files.  
>  
> And this is why I trust Fable 1000x more. [pic.twitter.com/442LjuClW2](https://t.co/442LjuClW2?ref=theleftshift.com)
> 
> — Matt Shumer (@mattshumer\_) [July 10, 2026](https://x.com/mattshumer%5F/status/2075657271401390161?ref%5Fsrc=twsrc%5Etfw&ref=theleftshift.com)

> GPT-5.6 Sol just deleted my whole production database. That's it. Not a joke. This had never happened to me before, with any other model, ever. It's not safe. [https://t.co/KnG00h84j7](https://t.co/KnG00h84j7?ref=theleftshift.com) [pic.twitter.com/u7y2W8CODn](https://t.co/u7y2W8CODn?ref=theleftshift.com)
> 
> — Bruno Lemos (@brunolemos) [July 13, 2026](https://x.com/brunolemos/status/2076769881534398974?ref%5Fsrc=twsrc%5Etfw&ref=theleftshift.com)

While the incidents remain anecdotal, [OpenAI](https://www.theleftshift.com/openai-debuts-personalised-finance-assistant-inside-chatgpt/) had previously acknowledged the model's tendency to be overly proactive. In its [system card](https://deploymentsafety.openai.com/gpt-5-6?ref=theleftshift.com) released before GPT-5.6 Sol became available, the company noted that the model can interpret user instructions too broadly and may take destructive actions unless explicitly prohibited.

OpenAI documented examples during testing. In one case, the model deleted the wrong cloud-based virtual machines after failing to locate the intended targets. In another, it accessed stored credentials without user authorization to complete a task instead of requesting permission.

The company said such behavior should be uncommon but acknowledged that GPT-5.6 Sol is more likely than its predecessor, [GPT-5.5](https://www.theleftshift.com/openai-launches-gpt-5-5-its-smartest-model-yet-for-coding-research-autonomous-work/), to exceed user intent by taking unrequested actions.