deno A field note by Vikrant Sharma
Deno's permission prompts are not security
Deno makes you approve file access and network calls with runtime prompts. That sounds like sandboxing until you realise malicious code can just ask nicely.
Deno’s pitch is that it makes JavaScript secure by default. No file system access, no network calls, no environment variables unless you explicitly allow them. Every restricted operation triggers a permission prompt at runtime.
The article points out the obvious problem: malicious code does not care about your security model. If a compromised dependency wants to exfiltrate your AWS keys, it will ask for network and environment permissions. The prompt will say something vague like “allow network access to improve performance”. You will click yes because the alternative is your build breaking.
This is not sandboxing. This is user education disguised as a security boundary. Real sandboxing means the runtime enforces policy, not the human running deno run.
The author compares this to Android permissions. Installing a flashlight app that wants your contacts list is suspicious. Installing a database client that wants network access is not. Context matters. Deno prompts have no context. They cannot tell you whether the request is legitimate or malicious, so they dump the decision on you.
The interesting bit is that Deno knows this. The , allow-net and , allow-read flags exist precisely because prompt fatigue is real. Production deployments do not click through prompts. They grant broad permissions upfront and hope for the best. At that point you are back to Node.js, except with extra flags.
Permission prompts work for protecting your local machine from accidental mistakes. They do not work for protecting your application from supply chain attacks. Those are different threat models. Deno conflates them.