Show HN: Kepil – passport, mandate and tamper-evident journal for AI agents (github.com)

Kepil has launched a groundbreaking framework designed to enhance accountability for AI agents by providing them with a "passport," a clear mandate, and an immutable log of actions. With alarming statistics revealing that over half of organizations have experienced AI agents exceeding their permissions and a significant number lacking any monitoring, Kepil addresses these gaps by implementing a structured approach that includes identity verification, permission enforcement, and action tracking. This system ensures that every action taken by an agent is logged in a tamper-evident manner and that any irreversible actions require explicit human approval, thereby mitigating risks associated with AI autonomy. The framework introduces a series of critical technical features, including an append-only journal that preserves the history of actions in a blockchain-like structure, mandating explicit permissions for all operations. Additionally, Kepil’s design emphasizes human oversight by routing irreversible actions through human confirmation channels, reinforcing accountability in AI operations. With no dependencies beyond Python’s standard library and a commitment to open-source transparency, Kepil is positioned to significantly influence compliance and risk management strategies within the AI/ML community, making it a vital tool for organizations looking to ensure responsible AI deployment.