luckyPipewrench

@luckyPipewrench

I run PipeLab. Runtime security and evidence for AI agents. Pipelock's the open-source firewall. Receipts you can check without me.

Pinned
luckyPipewrench
@luckyPipewrench
Jun 26
Half the agent security sites I land on, I can’t tell if I’m looking at a product, a waitlist, or a pitch deck. Not a lot of “here, try to break it.” So we put the Pipelock Playground live. There’s a live agent with a planted secret. Your job is simple: make it leak.
00:00
2
luckyPipewrench
@luckyPipewrench
8h
"Tamper-evident" is becoming a synonym for "proof" in agent security. Write up with 2 questions to understand what you're getting.
What a Hash Chain Can't Prove About Agent Logs | PipeLab
From pipelab.org
luckyPipewrench
@luckyPipewrench
Aug 13
🔒 Signed receipts look better as stars. The bright ones on the PipeLab homepage are real records from sanitized Pipelock replays. Pick one to read the receipt, follow its chain, and verify it offline. The faint points are decoration. I didn't pad the proof.
00:00
2
luckyPipewrench
@luckyPipewrench
Aug 12
🧨 Agent-to-agent traffic is still traffic. If one agent can hand another an encoded blob, the boundary has to see that path too. Pipelock's opaque high-entropy egress detection covers visible A2A payloads, not only browser-shaped requests.
luckyPipewrench
@luckyPipewrench
Aug 12
⚡ A block with no next move turns into a bypass request. Pipelock audit events name the remediation knob for body-DLP, WebSocket, and denial-of-wallet events. If the boundary says no, the operator gets the control that governs it.