Definition
Where brute force guesses passwords, credential stuffing uses pairs that actually exist. That means few attempts per account, which slips past defenses that look for repeated login failures. The raw material is credential lists gathered from past data breaches and phishing, traded on the dark web.
A login with stolen credentials looks, to the system, exactly like the legitimate user. It wasn't attributed to credential stuffing specifically, but at Change Healthcare in the US, a fraudulent login to a single Citrix remote-access account without multi-factor authentication was the entry point that halted roughly a third of US medical-claims processing for weeks (Critical Brief 083). There was no layer separating "knows the password" from "is the legitimate authorized party."
The basics of defense: multi-factor authentication, ending password reuse, checking against known-leaked passwords, and blocking traffic that tries many accounts in a short time.
Lemma implementation
Credential stuffing works because a password — "know it and you're in" — is treated as authority itself. For AI agents' payments and operations, Lemma adds a layer that decides permission by proof, not by password. In Trust402, an agent must show with a zero-knowledge proof that it holds the role and spending limit before a payment, and requests whose proofs fail stop before execution.
When AI agents call Lemma's API, each key is bound to a scope, and creating, changing, or deleting schemas and documents stays inside that scope.
Operations made through Lemma are pinned as an audit trail. When suspicious use of a key is found, what was read and changed with it can be traced afterwards from a record that can't be altered.