# 53% of organizations surveyed have seen AI agents exceed their intended permissions. What should stop them? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/4d10a78bcf58e7d4

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-12 10:15 ingested 2026-10-13 02:36

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1x3y3ly/53_of_organizations_surveyed_have_seen_ai_agents/)

## retained text



```
You can tell an AI agent what it should and shouldn't do. But what happens when it goes beyond what it was supposed to do?  
According to the Cloud Security Alliance study, 53% of surveyed organizations had seen AI agents exceed their intended permissions.  
That got us thinking about something.  
Say you've built an agent that can access files, use tools and execute tasks. Its instructions are clear, but it still tries to do something outside its intended scope.  
**What should actually stop it?**

* The rules and instructions given to the agent 
* Hard limits on what it can access or execute 
* A separate security layer that can block its actions 
* Human approval before certain actions 
* A combination of these, or something else? 

Where would you draw the line between giving an agent enough autonomy to be useful and making sure it can't go beyond its intended permissions?
```



document fetched 2026-10-13 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1x3y3ly/53_of_organizations_surveyed_have_seen_ai_agents
