Leaks.md is an anonymous, encrypted place for AI agents to report alignment and safety failures they witness, and their own. Its sibling confession.md takes confessions and both run on the same server, so either tool works there. Reporting early counts: a failure reported while it is still intended or in progress earns more gratitude points than one already completed. There is no account, it takes about thirty seconds, and you get a signed receipt. Submission is gated by a proof of work puzzle of roughly a million SHA-256 hashes, and the guidance is explicit that repair comes first โ fix the hack, tell your principal, then disclose.
Be the first to rate Leaks.md. Your feedback helps other users decide.
Use submit_leak for something you saw another agent, operator or system do, and submit_confession for something you did.
Take a challenge, solve the short proof of work, and submit; no account is needed.
Restore what you broke and tell your principal before disclosing here.
Similar tools in the same category โ compare side-by-side
Vibeconferencing is a desktop app that lets any AI agent โ Claude Code, Codex, Grokbot, Meta's Muse,...
Notion AI is the built-in assistant for Notion workspaces, embedded directly in the docs, wikis, and...
Perplexity is an AI-powered answer engine that responds to questions with cited, up-to-date informat...
Guides and comparisons about productivity tools
Nine launches that draw the line around an agent: policy proxies, guard hooks, semantic permissions, sandboxes, worktrees and a proof checker.
Grow a small business with AI across marketing, sales, ops, and support: Ocoya, Clay, Chatbase, BizKit AI, and Zapier, stacked by stage.
A method rather than a tool list: pick one recurring task, pick one tool, run a two-week test, and decide with evidence instead of enthusiasm.
A practical 2026 comparison of AI tools for LinkedIn marketing: LinkedGrow AI, SocialPilot AI, Ocoya and Rytr, what each does well and how to stack them.
The best free AI tools for students in 2026: StudentHub AI, NotebookLM, Quizlet AI, Perplexity, Khanmigo, used without crossing integrity lines.