TLDR: It’s way too hard to build a demonstrably secure ChatGPT wrappers, or MCP servers - which severely hampers their use and value to society. And - it’s a solvable problem…so what are we waiting for?

The world wants these little wrappers (and MCP servers). These little wrappers can be very useful, and are very easy to make. It’s a watershed moment.

In order to use them, we need to trust them with sensitive data (whether it’s your company’s code, or an outpouring to your journal). Currently, it’s very hard to make wrappers that feel sufficiently trustworthy. It’s the limiting reagent to wrapper usefulness.

As always, security is never absolute and always comes at some cost. But I believe it’s possible to enable these wrappers to prove:

Here’s what each part of the stack needs to prove (and how readily provable it is today)

Let’s examine each with more detail.

The front end must prove sensitive data is not being sent anywhere sus

<aside>

What we can do now: For web: open source your front end - or just don’t minify it. Watch network requests. For native apps: 🤷

</aside>

On web - open sourcing has a decent solution, but it’s not ideal. Commercially, it makes your app much easier to copy. Also - it’s not that easy to audit.

On native apps - you can also open source your code, but there’s currently no way of proving that what’s open sourced is what you’re downloading. That’s a pity - and something app store owners could easily fix. It’s possible (but hard) to inspect network requests (especially on apple).