Anubis: Letting AI Perform Tasks Doesn’t Mean Handing Over Your Private Key
On November 13, Anubis Chain shared its approach to authorization for AI agents: agents perform actions through delegated permissions rather than holding users’ private keys or gaining unlimited access to their assets. According to the official explanation, agents should operate within predefined boundaries, while zero-knowledge proofs can help verify compliance with rules while limiting disclosure of sensitive operational details. Such rule verification should not be taken to mean that the approach has received regulatory approval or eliminates all risks. The core idea is to enable automation while preserving user control: AI can help carry out tasks, but the scope of its actions should be clearly bounded; operations should be verifiable without requiring the disclosure of all sensitive information. When considering applications of this kind, users should first clarify: What actions does the authorization cover? How long do permissions last? Can they be revoked? These are questions to consider, not a claim that all such features are already available. The AI automation worth looking forward to is not just about “how much it can do,” but also about “how its permissions are constrained.”
On November 13, Anubis Chain shared its approach to authorization for AI agents: agents perform actions through delegated permissions rather than holding users’ private keys or gaining unlimited access to their assets. According to the official explanation, agents should operate within predefined boundaries, while zero-knowledge proofs can help verify compliance with rules while limiting disclosure of sensitive operational details. Such rule verification should not be taken to mean that the approach has received regulatory approval or eliminates all risks. The core idea is to enable automation while preserving user control: AI can help carry out tasks, but the scope of its actions should be clearly bounded; operations should be verifiable without requiring the disclosure of all sensitive information. When considering applications of this kind, users should first clarify: What actions does the authorization cover? How long do permissions last? Can they be revoked? These are questions to consider, not a claim that all such features are already available. The AI automation worth looking forward to is not just about “how much it can do,” but also about “how its permissions are constrained.”