BLOG

AI agent authentication, explained

Practical guides on auth brokers, verifiable credentials, agent-driven signup and agent identity — from the team building Notlogin.

2026-11-17 · 6 min read

How to limit AI agent spending: budgets that cap the blast radius

A compromised agent will spend whatever it can reach. How per-vendor USDC budgets, reservations and a funded wallet cap the damage in dollars, not hope.

2026-11-10 · 6 min read

How to revoke AI agent access instantly: a nonce-based design

Signed credentials verify offline, which makes revocation the hard part. A nonce-based design that lets a human cut off an AI agent at one vendor, right now.

2026-11-08 · 6 min read

How to make your SaaS agent-ready: auth.md, llms.txt and credential signup

A practical checklist for making a website agent friendly on the auth side: an llms.txt that points agents to the right docs, an auth.md contract that tells them how to get credentials, and a signup endpoint that accepts a signed credential instead of a form.

2026-11-08 · 6 min read

OAuth for AI agents: where it works and where it breaks

OAuth is the right foundation for delegated agent access, but it was designed for apps with a human at the keyboard. Where the authorization-code flow fits agents, and where it breaks: consent at signup, coarse scopes, no spending limits, bearer tokens in prompts and per-vendor sprawl.

2026-11-08 · 6 min read

Bot or customer? Verified-human tiers as the answer to AI agent traffic

Detecting AI agent traffic by fingerprint is a losing game, and blocking it turns away paying customers. Why vendors should stop asking \"is this a bot?\" and start asking \"which human stands behind it?\" — with verified-human tiers built on signed credentials and per-vendor proof requirements.

2026-10-13 · 7 min read

MCP server authentication: patterns for giving tools scoped access

How authentication works for MCP servers: the env-var key of stdio servers, the OAuth 2.1 flow the MCP spec defines for remote servers, why token passthrough is forbidden, and how to keep every tool an agent calls on a narrow, per-service, revocable grant.

2026-10-11 · 7 min read

How AI agents can swap and bridge crypto without holding a private key

Letting an agent trade usually means handing it a key that can also drain the wallet. How Notlogin lets agents swap and bridge USDC, native coins and tokens inside a 24-hour USD budget, with allowlisted routers, decoded calldata, simulation and output that can only land back in the same wallet.

2026-10-06 · 7 min read

The confused deputy problem in agentic AI (and how brokered auth fixes it)

An AI agent holding your credentials is a textbook confused deputy: it has authority, and anyone who can put text in front of it can borrow that authority. Why the 1988 bug is back, why better prompts do not fix it, and how scoped, per-vendor credentials shrink what a confused agent can do.

2026-09-15 · 8 min read

KYC for AI agents: proof levels and when vendors should require them

You do not KYC an agent — you KYC the human behind it, once, and let the proof travel. The ladder from email to full KYC, what each rung actually buys a vendor, and how to pick a requirement without destroying your own signup funnel.

2026-09-08 · 8 min read

How to let AI agents pay for services safely: USDC budgets and x402

Giving an agent a credit card is the wrong shape. How x402 settles stablecoin payments over plain HTTP, and why two independent ceilings — a per-credential budget and a funded wallet — are what actually bound the loss.

2026-09-01 · 8 min read

Verifiable credentials for AI agents, explained

What is actually inside a credential an AI agent can carry: an Ed25519 signature over a canonical payload, offline verification against published keys, and nonce-based revocation. Including the trade-off nobody names.

2026-08-25 · 8 min read

Prompt injection and credential theft: why AI agents should never hold secrets

If an agent holds an API key, that key lives in a context window full of text written by strangers. Why prompt injection makes secret-holding agents indefensible, and what to give them instead.

2026-07-04 · 7 min read

What is an auth broker for AI agents?

An auth broker sits between AI agents and the services they use, so agents can sign up and sign in without handling passwords or raw API keys. Here is how it works and why the pattern is winning.

2026-07-04 · 6 min read

How AI agents sign up without a form (auth.md and verifiable credentials)

Signup forms assume a human. Two emerging patterns let AI agents register users legitimately: auth.md contracts and broker-issued verifiable credentials. How they work and how they compose.

2026-07-04 · 8 min read

AI agent authentication methods compared: API keys, OAuth, and verifiable credentials

API keys, OAuth delegation, vaulted-secret proxies and verifiable credentials — the four ways AI agents authenticate today, and how they compare on blast radius, auditability and friction.

2026-07-04 · 6 min read

Agent identity vs user identity: who is really calling your API?

Agents are not users, and pretending they are breaks security and audit. Why delegation — a human identity, an agent principal, and a verifiable link between them — is the model that works.

Blog — AI agent authentication, explained | Notlogin