Five free tools for AI security test inputs, OWASP risk mapping, release plans, response packs and agent tool-call packs. Use them on the web or through MCP, skills and plugins, with no PromptBrake account or API key. MCP prepares artifacts; it does not run tests. Run tool-call packs with the free GitHub Action using staging dispatcher capture. Response packs need a PromptBrake runner. Captured calls do not prove backend effects or permission enforcement.
AI apps can pass normal QA and still leak data, follow injected instructions, or take unsafe actions. AI Release Assurance tests your real staging chatbot, agent, RAG system, or API before deployment. Run 138 adversarial checks across 18 risk categories in your own environment, compare releases against a baseline, and enforce an APPROVE, BLOCK, or REVIEW decision in CI—with redacted evidence showing exactly why.
