Internal Safety Collapse: Turning the LLM or an AI Agent into a sensitive data generator.

Introducing XSafeClaw: The Open-Source Agent Safety Platform from Fudan University

An open taxonomy and scoring framework for evaluating AI agent sandboxes: 7 defense layers, 7 threat categories, 3 evaluation dimensions, 27 "sandboxes" scored.

Practices, protocols, and skills for AI-driven software development. 18 skills + 1 Bash safety hook for Claude Code, Codex CLI, OpenCode, Cursor, Gemini CLI, Antigravity, and any agent supporting the Agent Skills standard.

Human-in-the-loop execution for LLM agents

Trust nothing. Ship safely. — Skeptical-reading and prompt-injection defense skill for AI agents. Provenance tagging, red-flag patterns, refusal templates, and a read-only injection auditor. MIT.

Security scanner for AI agent tool definitions

Claude Code agent-in-container orchestration and automation

Guardrails service for AI agents. Default-deny tool call evaluation with LLM safety analysis, priority-ordered decision matrix, and human-in-the-loop escalations. Session recording, behavioral analysis, MCP proxy, secret redaction, and real-time audit.

Runtime detector for reward hacking and misalignment in LLM agents (89.7% F1 on 5,391 trajectories).

🛡️ Safe AI Agents through Action Classifier

Audit log + guard for AI agents. Passive logging, human-in-the-loop approval for dangerous ops (rm, drop, transfer) via Telegram. Diary, daily digest, timeline UI. Cursor & MCP ready. Cloudflare Workers + Hono + D1.

Deterministic execution authorization for AI agents

Open Threat Classification (OTC) — 10 threat patterns for AI agent skills, MCP servers, and plugins. CC-BY-4.0.

29 free, open-source plugins for Claude Code & Cowork — Google Drive, WhatsApp, YouTube, WordPress, Apollo & more. Built on the SOSA™ security framework.

Deterministic Guardrails for AI Agents. Ark acts as a logic-based firewall, preventing unauthorized actions through a rigorous rule engine. Ensure your AI behaves exactly as intended.

tethered — Runtime network egress control for Python. One function call to restrict which hosts your code can connect to.

Guardrails for LLMs: detect and block hallucinated tool calls to improve safety and reliability.

Cycles budget and action guard for OpenClaw agents

Policy engine for AI agents — enforceable rules, risk limits, approval gates, obligation tracking, and violation detection. One .acon file. Rust core + MCP server.