Niki A. Niyikiza's article explores how "The AI hallucinated" could become a compelling legal defense. The core problem lies in attributing responsibility when an AI agent makes a mistake. Without a clear link between a human's authorization and the agent's actions, it's difficult to refute this defense. Current systems often lack independently verifiable authorization evidence.

The piece argues that current logging practices aren't enough. While logs show *what* happened, they don't always prove *who* authorized it, and under *what* constraints. When high-stakes actions occur, like financial transactions, explicit authorization is crucial. The article advocates for authorization records as first-class artifacts, like checks, that can be presented as proof.

Niyikiza proposes the use of Tenuo warrants, cryptographically signed authorization objects, to solve this problem. These warrants are holder-bound, meaning they are linked to a specific agent key. They also have attenable scope, so delegated authority can only be narrowed. This ensures non-repudiation, providing a verifiable delegation chain across multi-hop execution.

In essence, the article calls for a shift from session-based authorization to action-level authorization. Tenuo warrants provide the necessary cryptographic proof for actions. They provide independently verifiable authorization evidence that survives multi-hop delegation. This approach makes "the AI did it" defense less viable, enhancing accountability in AI-driven systems.