hkpatel3

To avoid browser pop-ups during provisioning and credential exchange, we recommend the following flow:

Sign in to your Stripe account. Associate your existing provider account (or create a new one) with stripe projects link. Add a payment method with stripe projects billing add. Start the agent session.

This is a note in their docs. I think there still needs a better way for autonomous agents.

I think there is still room, since, as I understand it, I do not want coding agents to use the same account that I end up using. I understand that some users may not care, but I think agents should have their own accounts and the ability to create their own accounts and use services like Vercel, Supabase, etc.

Openrouter can also be used with claude code. https://openrouter.ai/docs/guides/claude-code-integration

I have never tried to decompile using an LLM but I have heard that it can recognize the binary patterns and do it. Has anyone tried to decompile a major software and been successful?

Brave Article: https://brave.com/blog/comet-prompt-injection/

I developed Auth Agent which is an OpenID Connect Provider that gives your AI it's own credentials rather than using your credentials. It is open source. Add it as a sign in option to your websites now.

I would disagree since I believe web agents and computer using agents that act like human on the web do need a way to authenticate and UI has a big component to that. A2A works with implicit OAuth Flow but for web agents and computer using agents having a flow of OAuth 2.1 + OIDC is rational. Check this repo out once https://github.com/auth-agent/auth-agent. I do want to read the article fully and I will get back to you. Thanks for the article!

I created Autosign earlier, which combined AgentMail and BrowserUse to let agents pass email verification. It worked for a while but I still faced problems like hCaptcha and other blockers. That made it clear to me that AI agents need an efficient way to sign in that is not a turnaround or bypass solution built on top of human login systems.

Auth-Agent is the result of that. It is the world’s first OpenID Connect provider for AI agents. It provides a standard OAuth 2.1 / OIDC flow but the client is an AI agent instead of a human user. It is open source: https://github.com/auth-agent/auth-agent.

Your credentials can be easily stolen through prompt injection in Comet and ChatGPT Atlas. Auth-Agent, in simple words, is sign in with google but for ai agents (OIDC Provider + OAuth 2.1).

It allows agents like browser use, browserbase, ChatGPT Atlas, Comet, easily authenticate across websites. Essentially, made for web agents and computer using agents. Feel free to comment here or raise an issue in the Github repo.

Website coming soon.

Perplexity Comet and ChatGPT Atlas authenticate using human credentials, making them vulnerable to prompt injection attacks. Auth-Agent solves this by giving AI agents their own credentials, enabling secure authentication for web and computer-using agents.

The current solution has browser use as the example web agent and it can easily sign in with auth agent.

This is fully open source. Please take a look at the repo and feel free to raise any issue or let me know your thoughts.