Watchtower

Task-scoped runtime security for AI agents

1 follower

Task-scoped runtime security for AI agents

1 follower

Watchtower is an open-source experiment in task-scoped authority for autonomous AI agents. Define an agent's task and capabilities, generate an authority boundary, then test actions as ALLOW, REVIEW, or BLOCK. The MVP includes a Security Lab and an audit trail. What's different: we're exploring whether an agent's authority can be derived from its current task and eventually enforced at the runtime layer.

Free

Launch tags:Developer Tools•Artificial Intelligence

Launch Team

Dropbox SignSimple, secure eSignatures for the way your team works.

Promoted

Maker

📌

Hey Product Hunt 👋 I built Watchtower around one question: Should an AI agent's authority depend on the task it's currently performing? If an agent is asked to fix an authentication bug, it may need access to a workspace, Git and tests. But why should it also be able to read SSH keys, access production systems, or make payments? Watchtower's MVP lets you define an agent, generate a task-scoped authority boundary, and test actions with ALLOW / REVIEW / BLOCK decisions. It's deliberately early: the current web security decisions are simulated, and the Linux runtime is experimental. The long-term goal is to turn task-scoped authority into a real, enforceable runtime boundary. Try it and tell me what you think, especially what you would change.

Report

2mo ago

Reviews

No reviews yetBe the first to leave a review for Watchtower