You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$500 left to reach the funding goal.
Agent Circuit Breaker (ACB) is an open-source runtime safety layer for AI agents. It checks potentially dangerous actions before execution and can allow, block, or require human approval for operations involving shell commands, MCP tools, SQL, and filesystem access.
The project addresses a practical gap in agent safety: an AI model can produce an unsafe tool call even when the surrounding application is designed to behave responsibly. ACB provides deterministic checks that do not depend on asking another language model to judge whether an action is safe.
The goal is to make these controls accessible to developers building AI agents, including people who cannot afford proprietary security infrastructure.
The goal is to make deterministic runtime safety easier to integrate into real AI-agent applications.
I will continue developing ACB as an open-source Python project, improve its protections against dangerous tool actions, expand regression tests for known failure modes, and improve the documentation and integration examples.
I will prioritize verifiable behavior over claims of comprehensive agent security. ACB is a safety gate, not a substitute for operating-system isolation, least-privilege permissions, or other security controls.
Success will be measured through reproducible tests and practical adoption rather than claims that the tool makes agents completely safe.
Measures will include:
Regression tests demonstrating that defined dangerous operations are blocked or routed for approval.
Tests for bypass attempts, malformed inputs, and edge cases.
Clear documentation of supported behaviors, limitations, and deployment assumptions.
A straightforward integration path that developers can test locally.
Feedback from developers using ACB in real agent workflows.
I will publish test results and document any important limitations discovered during validation.
The requested $500 will support a focused development and validation milestone for this open-source project.
I plan to use the funding for the time and resources required to strengthen security tests, validate safety-critical behavior, improve developer documentation, and produce reproducible examples that demonstrate how ACB can be integrated into agent applications.
The work will remain publicly available so that developers can inspect, test, and build on the results.
I am Sagar Chhatrala, a security engineer working on cloud security, infrastructure security, and AI-agent runtime safety.
I created and maintain Agent Circuit Breaker as an independent open-source project. My professional experience includes cloud platforms, Kubernetes, identity and access management, security automation, and security controls for production systems. I have also contributed to cloud security benchmark work and spoken at CrowdStrike Fal.Con.
ACB builds on this background by applying established security engineering principles—explicit authorization, deterministic enforcement, auditability, and fail-safe handling—to AI-agent tool execution.
Project: https://github.com/sagarchhatrala/agent-circuit-breaker
The main risk is that deterministic checks cannot anticipate every dangerous action or every way an agent integration can be bypassed. Developers may also find that integrating a separate safety layer requires more work than expected.
If the project fails to gain adoption, the resulting tests, implementation, and documented limitations can still help other developers understand the strengths and weaknesses of deterministic runtime controls.
If validation reveals security gaps, I will document them rather than claim that ACB provides protection it has not demonstrated. The project will not be presented as a complete sandbox or as a guarantee against arbitrary-agent compromise.
0
There are no bids on this project.