Blog: Authentication Is Largely Solved.
- For decades we've treated authentication as a proxy for authorization.
- For humans, that's been good enough.
- A "do you trust this person in this context" has been a reasonable proxy.
- You have to trust the other entity in perpetuity in an open-ended way, but that was generally fine for real, situated humans.
- But it doesn't work for agents, and it shows that it was always just a proxy.
- In the pre-LLM world, the distance between authentication and authorization felt like a trivial semantic distinction.
- In this new post-LLM world, the fundamental difference between authentication and authorization feels like a vast chasm.