PitchHut logo
PIC Standard: AI Action Firewall
Stop prompt injection from triggering tools.
Pitch

PIC Standard adds machine-verifiable contracts to agent calls and actions. Before any high-impact tool call, the agent must submit an Action Proposal (schema + verifier). If trust/evidence is insufficient, it fails closed and blocks the action.

Resolvable SHA-256 evidence. Ships with a CLI and drop-in LangGraph + MCP integrations.

Incoming cryptographic signing and signed API attestations.

Description

PIC Standard: Provenance & Intent Contracts

The PIC Standard offers an open protocol designed for causal governance in Agentic AI, emphasizing the importance of ensuring trust and accountability in autonomous systems. This standard is particularly valuable for bridging the Causal Gap that occurs when high-impact actions are taken based on input from untrusted sources. Through the use of Provenance & Intent Contracts (PIC), it sets forth machine-verifiable contracts that connect the provenance of inputs to the potential impact of actions.

Key Features

  • Secure Alignment: Establishes contracts that ensure actions are based on reliable input information, effectively addressing risks associated with untrusted sources.
  • Flexible Integration: Easily integrates with existing tools using the LangGraph framework, enhancing safety during execution.
  • JSON Contract Model: Utilizes a structured format for action proposals that includes detailed provenance and intent information, facilitating clear communication between agents and tools.
  • Focus on High-Impact Domains: Designed for sectors such as SaaS, FinTech, and enterprise applications where actions have significant consequences.

Technical Overview

The flow of interaction follows this model:

graph TD
    A[Untrusted Input] --> B{AI Agent / Planner}
    C[Trusted Data/DB] --> B
    B --> D[Action Proposal JSON]
    D --> E[PIC Verifier Middleware]
    E --> F{Valid Contract?}
    F -- Yes --> G[Tool Executor]
    F -- No --> H[Blocked / Alert Log]

Usage Example

A typical example of an action proposal in a trusted environment includes:

{
  "name": "payments_send",
  "args": {
    "amount": 500,
    "__pic": {
      "protocol": "PIC/1.0",
      "intent": "Send payment",
      "impact": "money",
      "provenance": [{"id": "invoice_123", "trust": "trusted"}],
      "claims": [{"text": "Pay $500", "evidence": ["invoice_123"]}],
      "action": {"tool": "payments_send", "args": {"amount": 500}}
    }
  },
  "id": "tool_call_1"
}

Comparative Advantage

The PIC Standard distinguishes itself from existing frameworks by emphasizing business logic and side effects, thus providing an essential foundation for the operation of enterprise agents in a secure manner.

Community Participation

Driven by an open-source ethos, the PIC Standard encourages participation from security researchers, framework authors, and enterprise architects to enhance its robustness and applicability across diverse fields.

For more details and to get involved, visit the repository at PIC Standard GitHub.

0 comments

No comments yet.

Sign in to be the first to comment.