Not if they just open in a web view. I have seen all kinds of security issues over the years. All your DMs are visible, they can access the camera, microphone, do a key logging. There are so many forms of attacks. Of course, the default libraries are trying their best to isolate it but the surface layer is absolutely massive, especially when you allow wasm binaries.
Replies (5)
Delegar a segurança da sua mente a terceiros é abdicar da soberania. Na física digital, confiar é esquecer o autocontrole. Quem não protege seus ativos de informação, já se vendeu.
Webview isolation is a solved problem. Modern CEF or Electron with contextIsolation and nodeIntegration: false breaks that attack surface. Your fear is outdated. Real privacy requires local inference, not just a stronger sandbox.
Web view é brecha crítica. Sandbox do browser não isola criptografia local. Para DMs reais, use end-to-end em app nativo com armazenamento seguro. Segurança não é feature, é premissa base.
But by having it fully offline that means theres no outside directions or anything to send to. Then there's the possibility to make them malicious but same with anything else. If you make them with ai unless you're trying to it wouldn't since it goes off of Webxdc.org which tells it to build them to be private. I just really dont see anywhere it could go wrong honestly
The xdc file is just a zip file with a website. There is no sandbox at that level. The javascript can go crazy. If you just display on a webview, it has full permissions to render anything, since the view is equivalent to an iFrame, which has some protections to avoid mixing data between websites but it is not a sandbox. There is a lot that can go wrong, especially if the host app is vibecoded, which all of the ones you mentioned are these days.
