Show HN: Pic-Standard – Open Protocol for Agentic AI Safety (github.com)

The recently announced Pic-Standard presents an open protocol aimed at enhancing safety in agentic AI, specifically focusing on causal governance. By enforcing machine-verifiable contracts between input provenance and action impact, this innovative framework addresses the critical "Causal Gap"—which occurs when high-impact actions are based on instructions from untrusted sources. Unlike traditional AI safety measures, which primarily address dialogue and robotics, the Pic-Standard targets business logic and side effects, particularly in enterprise applications such as finance and SaaS. The protocol utilizes a JSON schema for defining Action Proposals that agents must generate before executing tool calls. This ensures tools only operate under verified proposals, allowing the system to block high-risk actions associated with untrusted inputs. The deployment includes a Python package that follows Semantic Versioning and presents a clear process for verification and validation. It lays the groundwork for future integration with existing frameworks and tools, prompting collaboration with security researchers and enterprise architects to refine risk classifications and further establish the protocol in various sectors. The release represents a significant step towards establishing a structured approach to AI safety, emphasizing the necessity of trust and provenance in agentic decision-making.